University of Texas at El Paso DigitalCommons@UTEP

Open Access Theses & Dissertations

2018-01-01 Effects Of Wait Times And Bridge Tolls On Personal Vehicle Border Crossings Into El Paso Omar Solis University of Texas at El Paso, [email protected]

Follow this and additional works at: https://digitalcommons.utep.edu/open_etd Part of the Economics Commons

Recommended Citation Solis, Omar, "Effects Of Wait Times And Bridge Tolls On Personal Vehicle Border Crossings Into El Paso" (2018). Open Access Theses & Dissertations. 16. https://digitalcommons.utep.edu/open_etd/16

This is brought to you for free and open access by DigitalCommons@UTEP. It has been accepted for inclusion in Open Access Theses & Dissertations by an authorized administrator of DigitalCommons@UTEP. For more information, please contact [email protected]. EFFECTS OF WAIT TIMES AND BRIDGE TOLLS ON PERSONAL VEHICLE BORDER CROSSINGS INTO EL PASO

OMAR SOLIS Master’s Program in Economics

APPROVED:

Thomas Fullerton Jr., Ph.D., Chair

Rigoberto Delgado, Ph.D.

Dennis Bixler-Márquez, Ph.D.

Charles Ambler, Ph.D. Dean of the Graduate School EFFECTS OF WAIT TIMES AND BRIDGE TOLLS ON PERSONAL VEHICLE BORDER CROSSINGS INTO EL PASO

by

OMAR SOLIS, B.S.

THESIS

Presented to the Faculty of the Graduate School of The University of Texas at El Paso

in Partial Fulfillment of the Requirements for the Degree of

MASTER OF SCIENCE

Department of Economics and Finance THE UNIVERSITY OF TEXAS AT EL PASO December 2018 Abstract

Wait times at international ports of entry (POE) increase overall travel costs and negatively affect establishments at the border by raising costs and reducing efficiency. This study employs linear transfer function analysis to model northbound personal vehicle traffic flows across three international bridges that link Ciudad Juarez, Chihuahua, Mexico and El Paso, Texas, USA. The analysis attempts to build upon prior regional studies that examine the effects of tolls, exchange rates, and economic conditions on cross-border traffic volumes. This is done by incorporating a fairly unique data set of wait times into the equation specifications. Empirical results confirm inverse relationships between traffic flows and tolls, as well as between tolls and wait time, at each bridge. Substitution effects between structures are also observed. For the tolled ports of entry, dollar-equivalent toll estimates of the wait time effects are approximated using bridge-specific coefficients. A regional average of the cost effect is used to calculate a similar estimate for the un-tolled bridge. As value of time measures, wait time toll equivalents are used to estimate time- related crossing costs at each POE. To further assess model reliabilities, out-of-sample simulations are also utilized. U-statistics and error-differential regressions are deployed as tools for comparing

LTF accuracies relative to those of random walk benchmarks.

iii Table of Contents

Abstract ...... iii

Table of Contents ...... iv

List of Tables ...... v

List of Figures ...... vi

Chapter 1: Introduction ...... 1

Chapter 2: Literature Review ...... 4

Chapter 3: Data and Methodology ...... 7

Chapter 4: Empirical Results ...... 25

Chapter 5: Comparative Simulation Results ...... 35

Chapter 6: Conclusion...... 50

References ...... 53

Data Appendix ...... 55

Vita 63

iv List of Tables

Table 1. Variables and units of measure ...... 14

Table 2. Hypothesized effects on bridge crossings ...... 14

Table 3. Paso del Norte northbound small vehicles (BCPN)...... 26

Table 4. Ysleta-Zaragoza northbound small vehicles (BCYN) ...... 28

Table 5. Bridge of the Americas northbound small vehicles (BCAN) ...... 31

Table 6. Paso del Norte northbound small vehicles U-statistics ...... 36

Table 7. Paso del Norte U-statistic decompositions ...... 37

Table 8. Paso del Norte Ashley et al. (1980) error-differential regression results ...... 38

Table 9. Ysleta-Zaragoza northbound small vehicles U-statistics ...... 40

Table 10. Ysleta-Zaragoza U-statistic decompositions ...... 41

Table 11. Ysleta-Zaragoza Ashley et al. (1980) error-differential regression results ...... 42

Table 12. Bridge of the Americas northbound small vehicles U-statistics ...... 44

Table 13. Bridge of the Americas U-statistic decompositions ...... 45

Table 14. Bridge of the Americas Ashley et al. (1980) error-differential regression results ...... 46

Table 15. U-statistic and Ashley et al. (1980) accuracy comparisons ...... 48

Table 16. USCPI, NEXR, and Northbound Border Crossings Data ...... 55

Table 17. Northbound Personal Vehicle Toll data...... 57

Table 18. WTPN, WTYN, and WTAN Data ...... 59

Table 19. CJIMX, ELPM, REXR, and MXIP Data ...... 61

v List of Figures

Figure 1. Paso del Norte: Time Series of Personal Vehicles and Wait Times ...... 9

Figure 2. Paso del Norte: Personal Vehicle Traffic Against Wait Times ...... 10

Figure 3. Ysleta-Zaragoza: Time Series of Personal Vehicles and Wait Times ...... 10

Figure 4. Ysleta-Zaragoza: Personal Vehicle Traffic Against Wait Times ...... 11

Figure 5. Bridge of the Americas: Time Series of Personal Vehicles and Wait Times ...... 11

Figure 6. Bridge of the Americas: Personal Vehicle Traffic Against Wait Times ...... 12

vi Chapter 1: Introduction

Cross-border transportation plays a major role in the economies of many United States border cities. One analysis determines that Mexican nationals directly support 10% of economic activity and employment in the Rio Grande Valley of south-central Texas (Ghaddar et al., 2004).

In the same study, survey respondents further indicate that crossing times of two hours or longer act as deterrents to cross-border travel. A separate effort for the San Diego area finds that decreasing border wait times increase firm productivity while reducing shipping costs (ERB,

2008). Shorter cross-border wait times also increase the volumes of personal expenditures on both sides of the border due to improved consumer time usage efficiency.

Large numbers of cross-border shoppers and other visitors transit the United States-Mexico border each year. More than 74 million personal vehicle border crossings from Mexico to the

United States are estimated to have occurred in 2015, with 12 million crossing through a port of entry (POE) located in El Paso (BTS, 2016). This analysis focuses on northbound personal vehicle crossings through El Paso POEs because of the availability of a unique dataset on personal vehicle wait times for this area. The sample period for this analysis corresponds to the interval for which border crossing wait time data are available, August 2010 to December 2016.

The cities of El Paso and Ciudad Juarez are connected by a network of four international bridges that span the Rio Grande River—the natural landmark that delineates the border between

Texas and Mexico. Of the four, tolls are charged to travelers moving into and out of the country on three of the bridges. The untolled bridge is the Bridge of the Americas and the three tolled

1 bridges are the Paso del Norte or Santa Fe Street Bridge, the Stanton Street Bridge, and the Ysleta or Zaragoza Street Bridge. Of the tolled arteries, the Paso del Norte and Ysleta-Zaragoza POEs allow all personal vehicle traffic while northbound crossing on the Stanton Street Bridge is only allowed for vehicles enrolled in the SENTRI program.

The effect of tolls on transportation demand is mostly inelastic (Matas & Raymond, 2003).

Previous research indicates that an inelastic response of cross-border traffic to bridge tolls is typical of southbound traffic through the three tolled POEs in the El Paso area (De Leon et al.,

2009). Northbound traffic across the same structures is also inelastic with respect to toll tariffs

(Fullerton et al., 2013).

Just as tolls can affect bridge traffic volumes, long delays can also discourage cross-border commuting. This analysis attempts to determine the effect of wait times on northbound passenger vehicle travel through regional POEs. It takes advantage of wait time data reported online by U.S.

Customs & Border Protection (CBP) to model tolled northbound personal vehicle crossings through the Paso del Norte and Ysleta-Zaragoza ports (CBP, 2017). The study also examines the effects of wait times on traffic on the un-tolled Bridge of the Americas. Furthermore, wait time toll equivalents are calculated as monetary estimates of the time cost imposed on commuters using each POE. Those values quantify the monetary toll effect of each additional one-minute delay in crossing each bridge. Those estimates are also used to identifying the time threshold at which passenger cars will be induced to select an alternate route for crossing the border.

2 The next chapter reviews the literature on cross-border traffic, bottlenecks at POEs, and bridge tolls. The subsequent chapter summarizes the data and methodology utilized. Empirical estimation and comparative simulation results are discussed next, followed by a conclusion.

3 Chapter 2: Literature Review

Economies on either side of the United States-Mexico border are interconnected in a variety of ways. Cross-border shopping is an especially important type of interaction between cities on opposite sides of the international boundary (Coronado & Phillips, 2007). Factors such as price and income differentials between neighboring countries contribute to large volumes of cross-border shopping trips (Baruca & Zolfagharian, 2013). Differences in laws, policies, and consumer regulations also motivate consumers to purchase goods and services from across the border every day.

Prior studies report that the purchasing power of the Mexican peso is one factor that affects traffic at POEs. Real exchange rate fluctuations may have different effects depending on the location being examined (Fullerton, 2000). In El Paso, declines in the purchasing power of the peso reduce northbound traffic at the Bridge of the Americas, where vehicular travel is most prevalent. The Paso del Norte crossing experiences an increase in northbound pedestrian traffic when the peso weakens. In the same study, variations in the exchange rate do not noticeably affect international movements at the Ysleta-Zaragoza POE during the sample period analyzed.

Because many border crossings are tolled, it is reasonable to expect that tolls are likely to affect cross-border traffic. Governments often use tolls to aid in financing the construction and maintenance of new infrastructure. In addition, tolls sometimes generate a positive externality by alleviating congestion on roadways (Matas & Raymond, 2003). Hirschman et al. (1995) finds that the median toll elasticity for automobile traffic is around -0.10 for bridges and tunnels inside New

York City. Demand tends to be more responsive to price when traffic can move quickly through

4 an alternative path. If the alternative path is gridlocked, the elasticity of demand is usually smaller.

Elasticities for roadways frequently travelled by tourists are generally smaller than elasticities for roads traveled by a more diverse population. Inelastic tourist responses may result because tolls charged are only a small portion of the complete cost of any given trip (Matas & Raymond, 2003).

De Leon et al. (2009) analyzes the effects of exchange rate fluctuations, real bridge tolls, and economic activity on southbound traffic volumes at POEs in El Paso, Texas. Real peso depreciation is associated with statistically significant declines in passenger vehicle crossings at the Ysleta-Zaragoza Bridge. Furthermore, parameter estimates indicate that tolls negatively impact traffic volumes while economic activity positively affects bridge traffic.

In a subsequent study examining the effects on northbound crossings, Fullerton et al.

(2013) finds that tolls are inversely related to automobile traffic across both the Paso Del Norte and the Ysleta-Zaragoza POEs. Increased economic activity on either side of the international border stimulates northbound crossings at these POEs, while declines in the purchasing power of the peso have the opposite effect. Price elasticities are estimated at -0.502 at the Ysleta-Zaragoza bridge and at -0.226 at the Paso Del Norte bridge, implying that small vehicle traffic is relatively non-responsive to fluctuations in the tolls charged.

Consumer demand for international travel between the United States and Mexico creates congestion, bottlenecks, and increased wait times at the POEs. In addition to the direct financial costs that cross-border travelers incur, consumers who traverse the international line are also subject to time costs. Administrative decisions undertaken immediately after 11 September 2001

5 created long delays at U.S. POEs, causing traffic flows to change substantially (Olmedo & Soden,

2005).

As a deterrent to cross-border travel, long wait times can negatively impact the economies of border regions (Del Castillo-Vera, 2009). One proposed approach to reducing border wait times is to increase the number of CBP inspection personnel at POEs. Roberts et al. (2014) determines that a single additional agent at each of 17 land passenger POEs included in the analysis results in

3.6 million dollars in additional U.S. output. In El Paso, Texas, the higher staffing levels are estimated to increase personal vehicle crossings by 212,113 above the 2012 level and contribute an estimated 4 million dollars per year in time saved due to lower border wait times.

Previous research examines the effects of real tolls, exchange rates, and aggregate economic activity on cross-border travel. Another strand of the literature quantifies the economic effects of border-crossing delays. This study attempts to bridge these two bodies of research and contribute to the existing literature by assessing the ways in which cross-border traffic volumes are impacted by border wait times, tolls, exchanges rates, and economic activity. Monthly frequency data from 2010 through 2016 are utilized to complete the analysis. The data and methods employed are described in the next section.

6 Chapter 3: Data and Methodology

This study uses data for three of the four international bridges located in the El Paso-Ciudad

Juarez borderplex. The Ysleta-Zaragoza bridge is the easternmost POE within the El Paso city limits. This tolled POE is isolated from the others and is popular with professionals who commute across the border for work. The Paso del Norte bridge is the other tolled border crossing and is positioned near the city center. This bridge is used almost exclusively for small vehicle and pedestrian traffic flows to the downtown regions on both sides of the border. Lastly, the Bridge of the Americas is an untolled POE located approximately 3 miles east of the Paso del Norte crossing (De Leon et al., 2009).

In 2013, CBP selected the City of El Paso, Texas for a pilot program that allows reimbursing this federal agency for a portion of the services it provides (CBP, 2013). Under this agreement, the City of El Paso pays CBP for any overtime required to more fully staff POE lanes during peak hours (TTI, 2017). The El Paso program is one of 5 efforts that were approved at that time. Each program responds to increased trade/travel volumes across POEs and examines the effects associated with specific services (such as locally financed paid overtime) on the US economy (CBP, 2013). The effect of this program is demonstrated as decreased wait times in

Figures 1, 3, and 5 despite increased traffic volumes at each bridge.

The sample period for this study is August 2010 to December 2016. Monthly bridge traffic data are collected from CBP reports (CBP, 2017). Northbound tolls at the Paso del Norte and

Ysleta-Zaragoza bridges are reported by Caminos y Puentes Federales de Ingresos Conexos

(CAPUFE, 2017). Nominal tolls generally remain fixed for 1 to 2 years following a price change.

7 Real tolls, of course, change more often. Because of this, and because tolls across the two ports are not always uniform (rates changes are not always adopted simultaneously), the real toll variable is averaged across both tolled bridges as shown in Equation (1).

푡표푙푙 푇 푟푡⁄ 1 푁퐸푋푅푡 푇푂퐿퐿푡 = ∑ (100 ∗ ) (1) 2 푈푆퐶푃퐼푡 푡=1

푇푂퐿퐿푡 is the averaged real toll for month 푡. 푡표푙푙푟푡 is the nominal toll charged at bridge 푟 during month 푡. 푁퐸푋푅푡 is the nominal exchange rate between the Mexican peso and U.S. dollar during month 푡 and 푈푆퐶푃퐼푡 is the U.S. consumer price index for the same period (FRED, 2017).

Wait times are gathered from CBP reports (CBP, 2017). Labor constraints allow data collection only on weekdays between the hours of 8am and 5pm. The data are averaged to produce a monthly average wait time variable that is measured in minutes.

Figures 1, 3, and 5 chart personal vehicle crossings and wait times for each of the international bridges during the sample period. Each graph demonstrates a reduction of wait times around the year 2013 that likely results from the overtime staffing agreement between CBP and the City of El Paso (CBP, 2013), as well as infrastructure improvements that reduced bottlenecks at the Bridge of the Americas in 2012 (GAO, 2013). All bridges experience increases in small vehicle traffic during the sample period as well as diminished wait times relative to the period when staffing was not supplemented by local overtime budgeting.

8 Vehicular crossings at each bridge are also are plotted against the wait times for each structure in Figures 2, 4, and 6. Negative correlations are noted throughout. The sample correlation coefficients are calculated for each crossing; they are -0.417 for the Paso del Norte bridge, -0.482 for the Ysleta-Zaragoza bridge, and -0.343 for the Bridge of the Americas. These values and the scatter plots provide preliminary evidence that northbound automobile traffic reacts to fluctuations in commuting delays at the POEs in question.

0.300 60

0.250 50

0.200 40

0.150 30

0.100 20 Wait Time Wait Time (minutes)

0.050 10 Personal Personal Vehicles (millions)

0.000 0

Feb-11 Feb-12 Feb-13 Feb-14 Feb-15 Feb-16

Aug-10 Aug-11 Aug-12 Aug-13 Aug-14 Aug-15 Aug-16

Nov-10 Nov-11 Nov-12 Nov-13 Nov-14 Nov-15 Nov-16

May-14 May-11 May-12 May-13 May-15 May-16

Northbound personal vehicle crossings Personal vehicle wait times

Figure 1. Paso del Norte: Time Series of Personal Vehicles and Wait Times

9 0.300

0.250

0.200

0.150

0.100

Personal Personal Vehicles (millions) 0.050

0.000 0 10 20 30 40 50 60 Wait Time (minutes)

Figure 2. Paso del Norte: Personal Vehicle Traffic Against Wait Times

0.350 60

0.300 50 0.250 40 0.200 30 0.150 20

0.100 Wait Time Wait Time (minutes) 10

Personal Personal Vehicles (millions) 0.050

0.000 0

Feb-11 Feb-12 Feb-13 Feb-14 Feb-15 Feb-16

Aug-10 Aug-11 Aug-12 Aug-13 Aug-14 Aug-15 Aug-16

Nov-10 Nov-11 Nov-12 Nov-13 Nov-14 Nov-15 Nov-16

May-14 May-11 May-12 May-13 May-15 May-16

Northbound personal vehicle crossings Personal vehicle wait times

Figure 3. Ysleta-Zaragoza: Time Series of Personal Vehicles and Wait Times

10 0.350

0.300

0.250

0.200

0.150

0.100

Personal Personal Vehicles (millions) 0.050

0.000 0 10 20 30 40 50 60 Wait Time (minutes)

Figure 4. Ysleta-Zaragoza: Personal Vehicle Traffic Against Wait Times

0.450 60 0.400 50 0.350 0.300 40 0.250 30 0.200 0.150 20

0.100 Wait Time (minutes) 10 Personal Personal Vehicles (millions) 0.050

0.000 0

Feb-11 Feb-12 Feb-13 Feb-14 Feb-15 Feb-16

Aug-10 Aug-11 Aug-12 Aug-13 Aug-14 Aug-15 Aug-16

Nov-10 Nov-11 Nov-12 Nov-13 Nov-14 Nov-15 Nov-16

May-14 May-11 May-12 May-13 May-15 May-16

Northbound personal vehicle crossings Personal vehicle wait times

Figure 5. Bridge of the Americas: Time Series of Personal Vehicles and Wait Times

11 0.450 0.400 0.350 0.300 0.250 0.200 0.150

0.100 Personal Personal Vehicles (millions) 0.050 0.000 0 10 20 30 40 50 60 Wait Time (minutes)

Figure 6. Bridge of the Americas: Personal Vehicle Traffic Against Wait Times

In addition to tolls, traffic volumes, and wait times, the sample data include IMMEX international manufacturing employment in Ciudad Juarez, non-agricultural employment in El

Paso, the Mexican industrial production index, and a real peso-to-dollar exchange rate. IMMEX employment measures regional business cycle conditions in Ciudad Juarez. The industrial production index is a macroeconomic indicator used to assess the economic well-being of Mexico.

Both IMMEX employment and industrial production index data are from Instituto Nacional de

Estadistica y Geografia (INEGI, 2017). El Paso non-agricultural employment measures fluctuations in the business cycle of El Paso. It is reported by the Bureau of Labor Statistics (BLS,

2017). The real exchange rate is calculated by the Border Region Modeling Project at the

University of Texas at El Paso (BRMP, 2017).

Data are differenced once to induce stationarity. Dickey-Fuller unit root tests are used to confirm stationarity. Traffic volumes at each POE are analyzed using linear transfer function

12 (LTF) analysis. Cross-correlation functions are generated to aid in specifying equation lag structures. Autocorrelation functions are used to determine whether any autoregressive and moving average terms are required for modeling non-random movements in the LTF residuals.

An LTF with two independent variables, 푥 and 푧, plus autoregressive and moving average components, is shown in Equation (2).

푝 푞 푛 푘

푦푡 = 휃0 + ∑ 휙푖푦푡−푖 + ∑ 휃푗훾푡−푗 + ∑ 퐴푎푥푡−푎 + ∑ 퐵푏푧푡−푏 + 훾푡 (2) 푖=1 푗=1 푎=1 푏=1

LTFs are estimated for northbound personal vehicle crossings across each bridge. The general implicit function used for each POE is shown in Equation (3). The algebraic signs below the equation indicate the hypothesized relationship between the independent variables and the dependent variable. Table 1 provides variable acronym definitions and units of measure. Table 2 summarizes the hypothesized relationships between explanatory variables and individual traffic flows.

퐵퐶 = 푓(푇푂퐿퐿 , 푊푇 , 퐸퐿푃푀 , 퐶퐽퐼푀푋 , 푀푋퐼푃 , 푅퐸푋푅 , 퐴푅 , 푀퐴 ) (3) 푡 푡−푖 푡−푗 푡−푘 푡−푚 푡−푛 푡−푢 푡−푝 푡−푞 (+/−) (+/−) (+) (+) (+) (? )

13 Table 1. Variables and units of measure

Unit of Acronym Definition Measure BCPN Northbound personal vehicle crossings: Paso del Norte millions BCYN Northbound personal vehicle crossings: Ysleta-Zaragoza millions BCAN Northbound personal vehicle crossings: Bridge of the Americas millions TOLL Average real toll dollars WTPN Northbound personal vehicle wait times at Paso del Norte minutes WTYN Northbound personal vehicle wait times at Ysleta-Zaragoza minutes WTAN Northbound personal vehicle wait times at Bridge of the Americas minutes ELPM Non-agricultural employment in El Paso thousands CJIMX IMMEX employment in Ciudad Juarez thousands MXIP Mexican industrial production index Jan 2016 = 100 REXR Real exchange rate Jan 2016 = 100

Table 2. Hypothesized effects on bridge crossings

TOLL WTPN WTYN WTAN ELPM CJIMX MXIP REXR BCPN (-) (-) (+) (+) (+) (+) (+) (?) BCYN (-) (+) (-) (+) (+) (+) (+) (?) BCAN (+) (+) (+) (-) (+) (+) (+) (?)

Increases in real tolls are expected to decrease traffic across tolled POEs (De Leon et al.,

2009; Fullerton et al., 2013), while increasing car volumes at the untolled substitute Bridge of the

Americas (Matas & Raymond, 2003). Similarly, because movement delays represent a time cost, wait times are theorized to reduce traffic at gridlocked bridges while increasing car volumes at the less congested POEs. Increases in employment and industrial growth are hypothesized to increase cross-border traffic. The effect of the real exchange rate on bridge usage is ambiguous. Prior research shows that a weakening peso negatively impacts travel by Mexican consumers into El

Paso, but also incentivizes El Paso residents to travel into Mexico (Fullerton, 2000).

14 De Leon et al. (2009) reports that real peso depreciation is associated with decreased volumes of southbound traffic across the Ysleta-Zaragoza bridge. A subsequent study of northbound traffic determines that tolls significantly reduce crossings at both Paso del Norte and

Ysleta-Zaragoza bridges (Fullerton et al., 2013). As the expected effect on border traffic in response to increasing wait times is also negative, the proposed outcome of the two variables imply that wait times can be measured as a fee separate from the actual toll charged at the border crossing.

The monetary value placed on an additional unit of time represents the marginal value of time (Small, 2013). Much transportation research analyzes the valuation of time. In a study on toll saturation, Hensher et al. (2016) explains that the value of travel time savings is equal to the ratio of the marginal utility of time to the marginal utility of travel cost. Earnhart (2004) compares data on the time and monetary costs affecting actual demand for a recreational good, quantified by visits to a lake, with survey data indicating hypothetical responses to those factors. The effect of time (opportunity) costs on the demand for recreation is denoted 훽표 and the effect of access fees on demand is denoted 훽푎. If time costs are properly measured in monetary terms, then 훽표/ 훽푎 =

1 so that both time-related and non-time-related components of the cost of travel to the lake generate the same marginal effect on demand. This approach provides a ratio by which to adjust the monetary valuation of time costs using access fees as a benchmark.

Other studies apply comparable techniques to quantify the cost of time using ordinary least squares. McConnell and Strand (1981) employ a system where the opportunity cost of time is some variable, 푘, multiplied with average income. The variable 푘 equals the ratio of the OLS coefficients for travel cost and travel time expressed as a proportion of income. A separate but

15 comparable study utilizes a log-log specification to study the marginal value of time through elasticities of demand for air travel (De Vany, 1974). Adjusting this technique for a linear model produces a marginal value of time also equal to the coefficient of travel time divided by the coefficient of price.

Small (2013) presents a technique for determining the value of time using implicit differentiation. For simplicity, only contemporaneous lags of the regressors are used in Equation

퐿푇퐹 (4). 푋̂푟푡 represents the estimated total crossings at POE 푟. 푇푂퐿퐿푡 is the average toll of both bridges, and 푊푇푟푡 is the wait time at each bridge.

퐿푇퐹 푋̂푟푡 = 훽0 + 훽1푇푂퐿퐿푡 + 훽2푊푇푟푡 (4)

The total differential of Equation (4) is shown as Equation (5). The value of time for bridge crossers can be computed as the change in tolls that would exactly offset the effect of a given change in wait times so that total bridge crossings would remain unchanged. The change in crossings is set equal to zero in Equation (6) which is then transformed into Equation (7).

̂퐿푇퐹 ̂퐿푇퐹 퐿푇퐹 ∂푋푟푡 ∂푋푟푡 퐿푇퐹 푑푋̂푟푡 = 푑푇푂퐿퐿푡 + 푑푊푇푟푡, 푎푛푑 푑푋̂푟푡 = 0, (5) ∂푇푂퐿퐿푡 ∂푊푇푟푡

퐿푇퐹 퐿푇퐹 ∂푋̂푟푡 ∂푋̂푟푡 0 = 푑푇푂퐿퐿푡 + 푑푊푇푟푡 (6) ∂푇푂퐿퐿푡 ∂푊푇푟푡

퐿푇퐹 퐿푇퐹 ∂푋̂푟푡 ∂푋̂푟푡 푑푇푂퐿퐿푡 = − 푑푊푇푟푡 (7) ∂푇푂퐿퐿푡 ∂푊푇푟푡

16 ̂퐿푇퐹 ∂푋푟푡 ⁄ 푑푇푂퐿퐿 ∂푊푇 훽 푡 = − 푟푡 = − 2 (8) 푑푊푇 ̂퐿푇퐹 훽 푟푡 ∂푋푟푡 ⁄ 1 ∂푇푂퐿퐿푡

Isolating the derivative of tolls with respect to wait time yields Equation (8). Small (2013) uses the absolute value of the resulting implicit differentiation ratio, 훽2/훽1, in describing the value of time. Here, Equation (8) indicates that a one-minute change in wait times is equal to an increase in tolls of a magnitude 훽2/훽1, defined as the ratio of the estimated 푇푂퐿퐿푡 and 푊푇푟푡 coefficients from each LTF equation. This is shown in Equation (9).

푑푇푂퐿퐿푡 (∆푊푇푟푡) = 푂푝푝표푟푡푢푛푖푡푦 퐶표푠푡 표푓 퐵푟푖푑푔푒 퐶푟표푠푠푖푛푔푟 (9) 푑푊푇푟푡

Though subject to transaction costs, commuters intent on using a specific border crossing can elect to use one of the alternative bridges to minimize their cost of crossing. South of the border, transactional costs are affected by factors such as the distance between the original and alternate crossings and the potential for congestion on roadways that link the two bridges. Longer- than-planned distances from the substitute POE to the intended destination in El Paso can also influence the selection of a crossing point. Considering these costs, rational commuters are expected to use their preferred bridge if the crossing cost at that bridge (the sum of the

푂푝푝표푟푡푢푛푖푡푦 퐶표푠푡 표푓 퐵푟푖푑푔푒 퐶푟표푠푠푖푛푔푟 and 푇푂퐿퐿푡) is below the crossing cost at the other structures plus any transactional costs, or total cost.

17 In a frictionless system which ignores transactional costs, the untolled Bridge of the

Americas is always the ideal primary crossing, but the 푂푝푝표푟푡푢푛푖푡푦 퐶표푠푡 표푓 퐵푟푖푑푔푒 퐶푟표푠푠푖푛푔푟 cannot be directly evaluated there because a toll is not available for monetizing the wait time.

Instead, the average of 푇푂퐿퐿푡 coefficients from the Paso del Norte and Ysleta-Zaragoza regressions is applied to generate a mean monetary cost effect for crossings within the borderplex area. This method requires the assumption that the own-price elasticity of Bridge of the Americas traffic, with respect to tolls, is equal to the average of the own-price elasticities at the other bridges should the untolled bridge charge a fare. The wait time required for crossing costs at the Bridge of the Americas to exceed those of either tolled crossing is calculated by dividing the nominal toll by the nominal 푂푝푝표푟푡푢푛푖푡푦 퐶표푠푡 표푓 퐵푟푖푑푔푒 퐶푟표푠푠푖푛푔푟 at the Bridge of the Americas.

To asses LTF equation reliability, a sequence of 24-month out-of-sample forecasts are generated for the period between January 2013 and December 2016. This type of forecast simulation employs an expanding sample of historical data. The first step of the simulation utilizes a historical sample ranging from August 2010 to December 2012 and generates monthly forecasts for the term spanning January 2013 to December 2014. In the second step, the historical sample expands by one month, now incorporating a bank of historical observations that stretch from

August 2010 to January 2013. The second step provides forecasts for the period between February

2013 and January 2015. This sample expansion, model re-estimation, and simulation process is repeated through the end of the sample period, resulting in 48 1-month ahead forecasts, 47 2-month ahead forecasts, 46 3-month ahead forecasts and so forth.

18 Forecast reliability is assessed by comparing out-of-sample forecast results from the LTF model with those from random walk benchmarks. Two types of random walk models are used in this study. The first is a simple random walk (RW). At crossing 푟 and time 푡, RW predictions,

푅푊 푋̂푟푡 , are equal to the actual number of northbound small vehicle crossings in the previous month, or 푌푟푡−1, as shown in Equation (10).

푅푊 푋̂푟푡 = 푌푟푡−1 (10)

The second benchmark is a random walk with drift (RWD). The processes for developing

RW and RWD estimates are similar, but RWD methods include a drift component to consider differences in the last and second-to-last months of historical data. In Equation (11), RWD

푅푊퐷 predictions are illustrated by the variable 푋̂푟푡 . The drift component is defined as (푌푟푡−1 −

푌푟푡−2).

푅푊퐷 푋̂푟푡 = 푌푟푡−1 + (푌푟푡−1 − 푌푟푡−2) (11)

Relative accuracy is evaluated via two approaches. U-statistics provide descriptive information for each set of forecasts (Pindyck & Rubinfeld, 1998). A formal regression-based F- test based on pairwise error-differentials is also deployed (Ashley et al., 1980). First, stepwise U- statistics and U-statistic decompositions are calculated for LTF, RW, and RWD forecasts at each international crossing. Accuracy is determined by comparing these values across the three models.

Then, error-differential regression analysis indicates if the differences between the LTF and benchmark forecasts are statistically significant.

19 푓 U-statistics for each step, 푠, of the rolling regression are defined in Equation (12). 푋푟푡 is the forecast value of northbound personal automobile crossings at bridge 푟 and time 푡 as calculated by LTF, RW, and RWD models. 푌푟푡 is the actual observed value.

2 √1 푇 푓 ⁄ ∑푡=1(푋푟푡 − 푌푟푡) 푠 푇 푈푟 = (12) 2 √1 ∑푇 푓 √1 ∑푇 ( )2 ⁄푇 푡=1(푋푟푡) + ⁄푇 푡=1 푌푟푡

U-statistic values are always between 0 and 1. When equal to zero, a U-statistic describes

푓 a perfect forecast where 푋푟푡 is equal to 푌푟푡. A value near 1 indicates poor forecasting performance.

U-statistic second-moments can be decomposed into three proportions of inequality, the sum of which is equal to 1. These are bias, variance (푣푎푟), and covariance (푐표푣푎푟) proportions, illustrated in Equations (13), (14), and (15).

̅푓 ̅ 2 푠 (푋푟 − 푌푟) 푏푖푎푠푟 = 2 (13) 1 ∑푇 푓 ⁄푇 푡=1(푋푟푡 − 푌푟푡)

The bias proportion is a measure of the degree by which the averages of forecast and

̅푓 ̅ observed values differ from each other. These are 푋푟 and 푌푟, respectively. As an indication of systematic error within the model, it is preferable that the value of bias be close to zero.

푓 2 푠 (𝜎̂푟 − 𝜎푟) 푣푎푟푟 = 2 (14) 1 ∑푇 푓 ⁄푇 푡=1(푋푟푡 − 푌푟푡)

20 Further measuring systematic error, the variance proportion reflects model ability to

푓 replicate the variability in the actual series. In equation (14), 𝜎̂푟 and 𝜎푟 describe the standard deviations from forecast and observed series. A large 푣푎푟 value indicates that variability is much larger in one series than in the other. Large values imply that a model is poorly specified.

( ) 푓 푠 2 1 − 𝜌푟 𝜎̂푟 𝜎푟 푐표푣푎푟푟 = 2 (15) 1 ∑푇 푓 ⁄푇 푡=1(푋푟푡 − 푌푟푡)

The covariance proportion is an assessment of unsystematic error. Equation (15) introduces 𝜌, the correlation coefficient between the forecast and actual series. It is defined in

Equation (16).

푇 1 푓 ̅푓 ̅ 𝜌푟 = 푓 ∑(푋푟푡 − 푋푟 )(푌푟푡 − 푌푟) (16) 𝜎̂푟 𝜎푟 푇 푡=1

Ideally, bias and variance proportions equal zero, while the covariance proportion equals 1. That represents a situation in which all forecast errors are exclusively due to random shocks (Pindyck

& Rubinfeld, 1998).

Following descriptive assessment with U-statistics, pairwise error-differential tests

(Ashley et al., 1980) determine if there is a significant difference between the mean squared errors of LTF and benchmark forecasts. This method involves regressions of error terms from LTF and benchmark forecasts for each step-length. As shown in Equation (17), forecast errors (푒푟푡) are

21 calculated by subtracting observed values from forecast estimates, 푋̂푟푡. The process applies to

LTF, RW, and RWD models at each of the three border crossings.

푒푟푡 = 푋̂푟푡 − 푌푟푡 (17)

Using errors corresponding to step 푠 of the out-of-sample simulations, two new series are generated. In Equation (18), LTF error is added to the corresponding benchmark error. In

퐵퐾 Equation (19), it is subtracted from the same benchmark error. 푒푟푡 denotes the error from a

퐿푇퐹 benchmark forecast, either RW or RWD, at time 푡. 푒푟푡 is the LTF forecast error.

푠 퐵퐾 퐿푇퐹 훴푡 = 푒푡 + 푒푡 (18)

푠 퐵퐾 퐿푇퐹 훥푡 = 푒푡 − 푒푡 (19)

푠 푠 With 훴푡 and 훥푡 as the dependent variable, error-differential tests are executed as shown in

Equations (20) and (21). In those equations, 훼1, 훼2, 훼3, and 훼4 are regression coefficients to be

푠 푠 푠 푠 estimated. 휇(훴푡 ) and 휇(훥푡) are the sample means of the 훴푡 and 훥푡 series at step 푠. 휂푡 and 휀푡 are stochastic error terms of the error-differential equation. Significant differences between LTF and benchmark errors are determined by examining stepwise error means in combination with computed t-statistics and F-statistics for Equations (20) and (21). Error means are calculated as shown in Equation (22).

푠 푠 푠 훴푡 = 훼1 + 훼2[훥푡 − 휇(훥푡)] + 휂푡 (20)

푠 푠 푠 훥푡 = 훼3 + 훼4[훴푡 − 휇(훴푡 )] + 휀푡 (21)

22 푇 1 푒̅푠 = ∑ 푒 (22) 푟 푇 푟푡 푡=1

The algebraic sign of the error mean dictates how error-differential tests are completed. If

LTF and benchmark error means are of opposing signs, Equation (20) is employed. Otherwise,

Equation (21) is utilized. The error mean also serves in differentiating between the forecast accuracy of benchmark and LTF models. If the error mean of the benchmark forecast is positive, a positive value of 훼1 or 훼3 suggests that the accuracy of the LTF model is superior to the benchmark model if the value of 훼2 or 훼4 is also positive at that step. Given positive values for

훼2 or 훼4, a negative benchmark error mean in conjunction with negative values for 훼1 or 훼3 also implies that LTF forecast accuracy is better (Fullerton & Kelley, 2008).

The benchmark error mean further governs how significant differences between LTF and benchmark accuracy are determined. Whenever the error mean is positive, opposing signs of the

훼 coefficients in either equation indicate that a one-sided t-test should be employed to verify the statistical significance of each coefficient. The same is true whenever the error mean is negative and both 훼 coefficients are either positive or negative. LTF forecast accuracy is significantly better than benchmark forecast accuracy if the coefficient favoring the benchmark model is statistically insignificant while the coefficient favoring the LTF model is statistically significant.

Alternately, if both 훼 coefficients concur on which set of forecasts is most accurate, a non- standard F-test is adopted which examines the hypothesis that 훼1 = 훼2 = 0 or that 훼3 = 훼4 = 0.

Contrary to usual F-tests, this 4-pronged test requires significance levels at less than half of those

23 found in F-distribution tables (Ashley et al., 1980). Rejecting the null hypothesis establishes that differences in forecast accuracy between LTF and benchmark models are statistically significant.

Empirical results are reported in the next two chapters. First, estimation results for LTF equations at each POE are discussed. Those results include the effect of wait times, tolls, and other variables on northbound personal vehicle volumes at each POE. The opportunity cost of crossing is also examined. The subsequent chapter discusses out-of-sample simulation results plus forecast accuracy comparisons between LTF, RW, and RWD models.

24 Chapter 4: Empirical Results

Estimation results for each POE are presented in Tables 3 through 5. All series are differenced prior to estimation to remove non-stationarity. Price elasticities of demand are included in the second-to-last row. Persons crossing into the United States from Ciudad Juarez face both tolls and time costs, both of which affect traffic volume across the area’s POEs. The change in tolls required to achieve the same reduction in traffic as a one-minute change in wait time is reported in the final row of each table as the wait time toll estimate. The tolled bridge LTF results are discussed first, followed by the results for the untolled bridge.

Estimation results for the LTF equation for northbound personal vehicle travel across the

Paso del Norte bridge are shown in Table 3. The magnitude of the toll coefficient implies that a price hike in real tolls diminishes the amount of traffic using the Paso del Norte bridge. A one- dollar increase in the average toll charged for northbound passage reduces monthly traffic by nearly 17 thousand monthly personal vehicles. Fullerton et al. (2013) also reports decreased personal vehicle crossings in response to toll hikes at the Paso del Norte POE, though the magnitude of that toll coefficient is larger. Including wait times as explanatory variables may help to buffer the effect of tolls on traffic volume.

25 Table 3. Paso del Norte northbound small vehicles (BCPN)

Variable Coefficient Std. Error t-statistic Probability Constant 9.86E-06 0.001375 0.007177 0.9943 BCPN(-1) -0.414681 0.097972 -4.232632 0.0001 TOLL(-1) -0.016767 0.033832 -0.495604 0.6221 WTPN(-2) -0.000641 0.000285 -2.251661 0.0284 WTYN(-1) 0.000766 0.000327 2.343822 0.0227 WTAN 0.001225 0.000407 3.010488 0.0039 CJIMX(-10) 0.000512 0.000671 0.763283 0.4486 ELPM(-10) 0.001226 0.000805 1.523140 0.1335 REXR(-1) -0.001147 0.000494 -2.321186 0.0240 MXIP 0.002182 0.000630 3.462960 0.0010 AR(2) -0.432013 0.133117 -3.245366 0.0020 R-squared 0.558165 Mean Dep. Var. 0.001035 Adj. R-Sq 0.477832 S.D. Dep. Var. 0.017175 Std. Err. Reg 0.012411 Akaike Info. Crit. -5.783266 Sum Sq. Resid 0.008471 Schwarz Info. Crit. -5.418324 Log-Likelihood 201.8478 F-statistic 6.948094 Durbin-Watson 1.680352 Prob(F-statistic) 0.000001 Paso del Norte northbound small vehicle toll elasticity -0.151 Paso del Norte northbound small vehicle wait time toll estimate $0.038 Sample Period: August 2010 – December 2016 Number of Observations: 66

All three wait time coefficients are statistically significant at the 5% level. Traffic across the Paso del Norte POE responds negatively to increased northbound wait times across the same

POE at a rate of 641 fewer personal vehicles per month for each additional minute of wait time.

Increased wait times at the other international crossings, all else equal, repel traffic to the Paso del

Norte POE.

On average, an additional minute of wait time at the far-east Ysleta-Zaragoza bridge increases personal vehicle traffic across the Paso del Norte bridge by 766 cars, implying that border crossers are willing to drive a substantial 13.5 miles to a different area of the city to avoid the time cost of waiting. Wait times across the Bridge of the Americas produce a similar effect on Paso del 26 Norte traffic. An additional minute of wait time at the Bridge of the Americas is associated with an additional 1,225 personal vehicles across Paso del Norte. The somewhat larger effect probably results from the shorter distance between the Bridge of the Americas and the Paso del Norte bridge, which are less than 3 highway miles apart.

IMMEX employment in Ciudad Juarez and El Paso non-agricultural employment are positively correlated with northbound traffic across the Paso del Norte bridge. An increase of

1,000 in IMMEX and El Paso non-agricultural employment increases traffic by 512 and 1,226 personal vehicles, respectively. The 푅퐸푋푅 parameter indicates that decreases in the purchasing power of the peso reduce the number of monthly personal vehicles. A one percent depreciation of the peso leads to a 1.15 thousand decline in the number of cars traversing this bridge, similar to what is documented in Fullerton et al. (2013). 푀푋퐼푃 shows that a one-point increase in Mexican industrial activity raises the monthly traffic volume by nearly 2.2 thousand vehicles.

The estimated price elasticity of demand is -0.151. The coefficient for average tolls is not statistically significant, but the magnitude is comparable to that reported in Fullerton et al. (2013) for the same POE. Furthermore, an elasticity of -0.151 is comparable to elasticities described in the literature on toll road elasticity outlined by Matas and Raymond (2003). The estimate for the value of time spent on the bridge is included in the final row of Table 3. As expressed in Equation

(8) above, it is calculated as the absolute value of the quotient of the 푊푇푃푁 and 푇푂퐿퐿 coefficients at this bridge, or 훽푤푎푖푡 / 훽푡표푙푙 . Consumers traveling northward across the Paso del Norte POE are estimated to value their wait time at $0.04 per minute. Increasing the wait time at this POE by 1

27 hour relative to the other POEs has the same negative effect on personal vehicle cross-border activity at the bridge as adding $2.40 to the toll, or $2.50, in 2018 nominal terms.

Table 4. Ysleta-Zaragoza northbound small vehicles (BCYN)

Variable Coefficient Std. Error t-statistic Probability Constant 0.002924 0.001696 1.723500 0.0900 BCYN(-1) -0.493328 0.106128 -4.648422 0.0000 BCYN(-2) -0.473362 0.103641 -4.567327 0.0000 TOLL(-1) -0.031899 0.045744 -0.697328 0.4883 WTPN 0.000531 0.000340 1.563390 0.1233 WTYN(-2) -0.001266 0.000508 -2.491280 0.0156 WTAN(-2) 0.001348 0.000628 2.145338 0.0361 CJIMX(-5) 0.000465 0.000840 0.553600 0.5819 ELPM(-5) 0.001982 0.000969 2.045563 0.0453 REXR(-1) -0.001045 0.000589 -1.775486 0.0810 MXIP 0.001999 0.000813 2.459597 0.0169 AR(3) -0.476038 0.125994 -3.778254 0.0004 R-squared 0.509385 Mean Dep. Var. 0.002358 Adj. R-Sq. 0.417914 S.D. Dep. Var. 0.020743 Std. Err. Reg. 0.015826 Akaike Info. Crit. -5.290642 Sum Sq. Resid 0.014776 Schwarz Info. Crit. -4.908217 Log-Likelihood 199.8178 F-statistic 5.568838 Durbin-Watson 1.849027 Prob(F-statistic) 0.000005 Ysleta-Zaragoza northbound small vehicle toll elasticity -0.251 Ysleta-Zaragoza northbound small vehicle wait time toll estimate $0.040 Sample Period: August 2010 – December 2016 Number of Observations: 71

Regression results for northbound personal vehicle crossings across the Ysleta-

Zaragoza bridge are summarized in Table 4. Higher tolls also decrease northbound personal vehicle traffic volumes through the Ysleta-Zaragoza POE. A one-dollar increase in tolls reduces monthly northbound traffic by nearly 32 thousand personal vehicles. The larger negative toll elasticity at Ysleta-Zaragoza than at Paso del Norte is similar to that reported in Fullerton et al.

(2013). 28 All wait time coefficients at this bridge satisfy the 5% significance criterion. Congestion related substitution effects are also confirmed for this parabolic span. A one-minute increase in the wait times at the Ysleta-Zaragoza POE reduces monthly traffic across this bridge by 1,266 personal vehicles. This is the largest effect of own wait times on traffic across any of the three bridges and is consistent with the magnitude of the Ysleta-Zaragoza correlation coefficient discussed in conjunction with Figure 4.

Each minute of increased wait time at Paso del Norte increases personal vehicle traffic at

Ysleta-Zaragoza by 531. The effect of wait times at the Bridge of the Americas on traffic at the easternmost POE is more dramatic. An additional one-minute delay `at the Bridge of the Americas increases traffic at the Ysleta-Zaragoza crossing by 1,348 cars per month. The larger 푊푇퐴푁 coefficient may result from the closer proximity and easier access from the Bridge of the Americas to the Ysleta-Zaragoza bridge than from the Paso del Norte structure.

Coefficients for IMMEX employment in Ciudad Juarez and El Paso non-agricultural employment confirm positive relationships with northbound traffic. An increase of 1,000 in

Ciudad Juarez IMMEX employment increases monthly traffic across Ysleta-Zaragoza by 465 personal vehicles. The same increase in El Paso non-agricultural employment increases northbound personal vehicle travel by 1,982.

As the easternmost POE into El Paso, the Ysleta-Zaragoza bridge links Mexican consumers to large shopping centers on the east side of the city (De Leon et al., 2009). That geographic location makes the negative response of passenger vehicle crossings to real exchange rate, 푅퐸푋푅,

29 movements plausible and logical due to the consequent loss of purchasing power on the north side of the border (Fullerton et al., 2013). Mexican industrial production, 푀푋퐼푃, is positively correlated with northbound traffic across the Ysleta-Zaragoza bridge. One-unit increases in the

Mexican industrial production index are associated with an increase of 1,999 personal vehicles per month.

The price elasticity of demand at the Ysleta-Zaragoza crossing is also inelastic at -0.251.

As in Table 3, the coefficient for average tolls is not significant, but the value of the elasticity at

Ysleta-Zaragoza is akin to that reported in Fullerton et al. (2013) and is also within the range outlined by Matas and Raymond (2003). Cross-border commuters at the Ysleta-Zaragoza bridge value their time at $0.04 per minute, as well. For this bridge, the value of time spent waiting in line to cross is estimated by dividing the 푊푇푌푁 coefficient by the Ysleta-Zaragoza 푇푂퐿퐿 coefficient, or 훽푤푎푖푡 / 훽푡표푙푙. The absolute value is equal to $0.04. Equal time valuations between

Paso del Norte and Ysleta-Zaragoza bridges imply that both tolled POEs are indistinguishable in terms of utility provided to persons engaging in northbound passage into the United States.

In 2018 nominal terms, the average toll is equal to $1.46 and the per hour wait time toll estimate is $2.50 at both tolled crossings. If a commuter prefers a tolled POE currently gridlocked with a one-hour wait, the total cost at the other POE must be less than $3.96 for the same commuter to consider using the secondary crossing. The geospatial location of primary and secondary selections will influence the extent of transactional costs, but it is possible that smaller wait times at the other crossing will reduce total costs by a large enough margin to incentivize using the alternative bridge.

30 Table 5 summarizes the LTF estimation results for northbound traffic across the untolled

Bridge of the Americas. The effects of increasing tolls are as expected. The toll coefficient indicates that toll hikes at the neighboring bridges result in significantly more traffic through the

Bridge of the Americas at approximately 106.5 thousand additional personal vehicles per month for each dollar increase.

Table 5. Bridge of the Americas northbound small vehicles (BCAN) Variable Coefficient Std. Error t-statistic Probability Constant -0.000887 0.001637 -0.541615 0.5902 BCAN(-1) -0.650255 0.079283 -8.201658 0.0000 TOLL 0.106597 0.042727 2.494832 0.0155 WTPN 0.000613 0.000400 1.533126 0.1308 WTYN(-1) 0.001049 0.000563 1.864895 0.0673 WTAN(-1) -0.000731 0.000660 -1.107421 0.2728 CJIMX(-7) 0.002216 0.000986 2.248942 0.0284 ELPM(-1) 0.002060 0.000952 2.164176 0.0347 REXR(-1) -0.001092 0.000563 -1.938507 0.0575 MXIP 0.001937 0.000822 2.356601 0.0219 AR(2) -0.391462 0.134088 -2.919434 0.0050 MA(6) -0.319493 0.155444 -2.055351 0.0444 R-squared 0.611453 Mean Dep. Var. 0.000388 Adj. R-Sq. 0.536471 S.D. Dep. Var. 0.024707 Std. Err. Reg. 0.016821 Akaike Info. Crit. -5.160283 Sum Sq. Resid 0.016129 Schwarz Info. Crit. -4.771743 Log-Likelihood 190.0298 F-statistic 8.154598 Durbin-Watson 1.815802 Prob(F-statistic) 0.000000 Bridge of the Americas northbound small vehicle toll elasticity 0.652 Bridge of the Americas northbound small vehicle wait time toll estimate $0.030 Sample Period: August 2010 – December 2016 Number of Observations: 69

All three wait time coefficients exhibit the hypothesized signs, albeit with computed t- statistics that do not satisfy the standard significance criterion. As wait times at the Bridge of the

Americas increase by one-minute, total monthly traffic across the same POE decreases by 731

31 cars. A one-minute increase in the average wait times at the Paso del Norte crossing is shown to increase monthly traffic across the Bridge of the Americas by 613 personal vehicles. Likewise, a one-minute increase in the wait time for personal vehicles at Ysleta-Zaragoza is expected to increase northbound personal vehicle crossings at the Bridge of the Americas by 1,049 automobiles per month.

Changes in 퐶퐽퐼푀푋 and 퐸퐿푀푃 are positively correlated with traffic volumes at the untolled

Bridge of the Americas. As IMMEX employment in Ciudad Juarez grows by 1,000 employees, monthly personal vehicle traffic across the Bridge of the Americas is expected to increase by 2,216.

An increase of 1,000 in El Paso non-agricultural employment raises monthly traffic flows across the bridge by 2,060 personal vehicles.

The coefficient for the real exchange rate is negative. Fullerton (2000) finds that an increase in the real exchange rate decreases the amount of total traffic through the free POE. These results corroborate those findings. Personal vehicle traffic over the Bridge of the Americas responds to decreases in the valuation of the peso in the same way that personal vehicle traffic responds at the Paso del Norte and Ysleta-Zaragoza bridges. The similarity in magnitude of the

푅퐸푋푅 coefficients suggest that northbound travel into El Paso is reduced uniformly across all three

POEs whenever the peso weakens.

The estimated coefficient for 푀푋퐼푃 is also positive in Table 5. As in Tables 3 and 4, it is also statistically significant. As 푀푋퐼푃 goes up by one unit, personal vehicle traffic across the

Bridge of the Americas will increase by 1,937 personal vehicles per month.

32 Bridge of the Americas small vehicle toll elasticity is calculated using 푇푂퐿퐿푡 and 퐵퐶퐴푁 series averages along with the regression coefficient for 푇푂퐿퐿. At 0.652, the magnitude of the toll elasticity is larger than those calculated for the Paso del Norte or Ysleta-Zaragoza bridges. While still falling within the inelastic range, it suggests that northbound traffic across the un-tolled Bridge of the Americas is more responsive to toll hikes than the traffic flows at either of the tolled POEs.

As indicated by the positive sign of the parameter estimate for the 푇푂퐿퐿 variable, the Bridge of the Americas functions as a substitute route for the other bridges.

The absence of a toll prevents direct calculation of a wait time toll equivalent at this bridge.

To circumvent that, a regional average toll effect is generated using the 푇푂퐿퐿 coefficients reported in the Paso del Norte and Ysleta-Zaragoza tables. The Small (2013) method is slightly modified such that the 푊푇퐴푁 coefficient is divided by the average of the 푇푂퐿퐿 coefficients from the two other bridges, or 2 ∗ 훽푤푎푖푡 / (훽푡표푙푙푃퐷푁 + 훽푡표푙푙푌푍). Numerically, the opportunity cost of crossing at the Bridge of the Americas is determined with the absolute value of the following: 2 ∗

−0.000731 / (−0.016767 − 0.031899) = $0.03.

On average, the Bridge of the Americas experiences a larger number of personal vehicle crossings than either of the tolled bridges. This effect is likely due to the absence of a toll. At

$0.03 a minute, the wait time toll equivalent is smaller than that of the Paso del Norte or Ysleta-

Zaragoza crossings. Commuters utilizing this bridge as their primary selection are expected to continue using it until the crossing cost exceeds the total cost at either of the alternatives. Because it is untolled, the Bridge of the Americas can absorb a greater wait time cost than the other bridges before commuters employ alternative routes, depending on transactional costs. In a frictionless

33 system, that exchange occurs when the wait time cost at this crossing is greater than 푇푂퐿퐿푡. In

2018, that is expected to occur after approximately 49 minutes.

Out-of-sample simulation accuracy of the LTF models is assessed in the next chapter.

Random walk benchmarks for each bridge are used for that exercise. The accuracy analysis attempts to determine whether the LTF models provide more accurate traffic volume forecasts than the random walks. Thiel U-coefficients are generated for LTF forecast errors at each step-length for 1-month-ahead all the way to 24-months-ahead. The reliability of the LTF models is then assessed by comparing those U-statistics against the values calculated for the random walks

(Pindyck & Rubinfeld, 1998). A second approach is utilized wherein error-differentials between

LTF and random walk forecasts are included in a regression equation to determine if any difference in the accuracy of the two approaches is statistically significant (Ashley et al., 1980).

34 Chapter 5: Comparative Simulation Results

Out-of-sample forecast accuracy is examined in this chapter. Using an expanding window of historical observations, the forecast errors of LTF models are compared to those of RW and

RWD benchmarks. Descriptive U-statistics determine comparative accuracy rankings among the three types of forecasts at each POE. Those results are then supplemented with outcomes of the

Ashley et al. (1980) error-differential regression technique to establish whether pairwise differences in precision between LTF and RW or LTF and RWD models are statistically significant.

Paso del Norte simulation accuracy results appear in Tables 6 through 8, followed by results from the Ysleta-Zaragoza crossing in Tables 9 through 11. The Bridge of the Americas outcomes are summarized in Tables 12 through 14. Of the three tables assigned to each POE, the first compares U-statistics between LTF, RW, and RWD models throughout the steps of the simulation. The second table shows U-statistic second-moment decompositions. Results from

Ashley et al. (1980) error-differential tests are shown in the last of the tables for each bridge. The final table in this chapter, Table 15, compares accuracy from U-statistic and error-differential analyses across the three sets of forecasts for each of the international bridges.

Table 6 summarizes U-statistics from LTF and benchmark forecasts at the Paso del Norte bridge. Figures in bold typeface highlight the smallest U-statistic, which represents the most accurate reading at that step (Pindyck & Rubinfeld, 1998). U-statistics show that the LTF approach is most accurate at step-lengths 1 through 11. At steps 7 and 8, the RW U-statistics are equal in magnitude to those of the LTF approach. After the 11-month step-length, all LTF U-statistics are

35 larger than corresponding values for RW forecasts. Both LTF and RW U-statistics are consistently smaller than any from the RWD model at those longer step-lengths.

Table 6. Paso del Norte northbound small vehicles U-statistics

Step- Observations LTF RW RWD Length 1-Month 48 0.026 0.042 0.071 2-Month 47 0.032 0.042 0.069 3-Month 46 0.034 0.042 0.069 4-Month 45 0.035 0.041 0.069 5-Month 44 0.037 0.041 0.068 6-Month 43 0.040 0.041 0.068 7-Month 42 0.042 0.042 0.068 8-Month 41 0.042 0.042 0.069 9-Month 40 0.040 0.042 0.069 10-Month 39 0.040 0.042 0.070 11-Month 38 0.042 0.043 0.070 12-Month 37 0.046 0.043 0.071 13-Month 36 0.049 0.043 0.071 14-Month 35 0.050 0.043 0.071 15-Month 34 0.050 0.043 0.071 16-Month 33 0.050 0.039 0.067 17-Month 32 0.049 0.040 0.064 18-Month 31 0.048 0.039 0.064 19-Month 30 0.045 0.038 0.062 20-Month 29 0.042 0.039 0.062 21-Month 28 0.042 0.039 0.062 22-Month 27 0.046 0.040 0.063 23-Month 26 0.050 0.039 0.063 24-Month 25 0.053 0.040 0.063

Table 7 describes the decompositions for each Paso del Norte U-statistic in Table 6. Bias and variance components indicate systematic forecast errors. The covariance proportion measures unsystematic errors. If forecast errors are entirely random, the sum of bias and variance components equals 0 and the covariance component is equal to 1. The sum of the three components always equals 1 (Pindyck & Rubinfeld, 1998). For all three approaches, the greatest source of error is random at each step-length.

36 Table 7. Paso del Norte U-statistic decompositions

LTF RW RWD Step- Bias Var Covar Bias Var Covar Bias Var Covar Length 1-Month 0.001 0.006 0.993 0.003 0.000 0.997 0.000 0.125 0.875 2-Month 0.003 0.001 0.996 0.008 0.002 0.990 0.002 0.160 0.838 3-Month 0.005 0.010 0.986 0.010 0.006 0.984 0.001 0.159 0.840 4-Month 0.007 0.020 0.973 0.005 0.000 0.995 0.000 0.137 0.863 5-Month 0.013 0.037 0.951 0.008 0.002 0.990 0.001 0.176 0.822 6-Month 0.012 0.035 0.952 0.005 0.000 0.995 0.000 0.159 0.841 7-Month 0.009 0.044 0.947 0.004 0.000 0.995 0.001 0.173 0.826 8-Month 0.008 0.047 0.944 0.003 0.000 0.997 0.000 0.171 0.829 9-Month 0.011 0.069 0.920 0.003 0.000 0.997 0.001 0.177 0.822 10-Month 0.015 0.097 0.889 0.005 0.000 0.995 0.001 0.193 0.806 11-Month 0.017 0.109 0.874 0.003 0.000 0.997 0.000 0.183 0.816 12-Month 0.015 0.115 0.870 0.004 0.000 0.995 0.001 0.201 0.798 13-Month 0.013 0.099 0.888 0.002 0.000 0.998 0.000 0.192 0.808 14-Month 0.017 0.103 0.880 0.005 0.001 0.995 0.002 0.219 0.779 15-Month 0.030 0.128 0.842 0.012 0.016 0.973 0.003 0.271 0.726 16-Month 0.039 0.120 0.840 0.002 0.000 0.998 0.000 0.215 0.785 17-Month 0.052 0.081 0.867 0.003 0.000 0.997 0.002 0.222 0.776 18-Month 0.058 0.057 0.886 0.000 0.000 1.000 0.000 0.224 0.775 19-Month 0.083 0.024 0.893 0.003 0.000 0.997 0.005 0.227 0.768 20-Month 0.131 0.011 0.858 0.002 0.000 0.998 0.001 0.211 0.788 21-Month 0.169 0.005 0.826 0.003 0.000 0.997 0.002 0.216 0.782 22-Month 0.175 0.014 0.810 0.006 0.000 0.994 0.003 0.237 0.759 23-Month 0.152 0.013 0.835 0.001 0.000 0.999 0.000 0.223 0.777 24-Month 0.131 0.017 0.852 0.002 0.000 0.998 0.003 0.221 0.777

Error-differential regression results for the Paso del Norte forecasts are summarized in

Table 8. Asterisks are used to highlight statistical significance. Rejection of the null hypothesis implies that LTF forecasts are significantly more accurate than RW or RWD forecasts at that step

(Ashley, et al., 1980).

37 Table 8. Paso del Norte Ashley et al. (1980) error-differential regression results

Step- Benchmark Error 훼 or 훼 훼 or 훼 1 3 2 4 F-statistic Conclusion† Length Forecast Mean t-statistic t-statistic 1-Month RW Negative -0.341 4.488* 10.130* Reject RWD Negative 0.072 8.671* 37.599* Reject 2-Month RW Negative -0.345 1.994* 2.047 Fail to Reject RWD Negative -0.150 5.807* 16.874* Reject 3-Month RW Negative -0.273 1.406* 1.025 Fail to Reject RWD Negative 0.035 5.140* 13.210* Reject 4-Month RW Negative 0.006 1.086 0.590 Fail to Reject RWD Negative 0.276 4.815* 11.631* Reject 5-Month RW Negative 0.088 0.727 0.268 Fail to Reject RWD Negative 0.183 4.259* 9.085* Reject 6-Month RW Negative 0.226 0.217 0.049 Fail to Reject RWD Negative 0.343 3.623* 6.624* Reject 7-Month RW Negative 0.140 -0.043 0.011 Fail to Reject RWD Negative 0.197 3.273* 5.376* Reject 8-Month RW Negative 0.206 0.026 0.021 Fail to Reject RWD Negative 0.239 3.329* 5.570* Reject 9-Month RW Negative 0.237 0.378 0.099 Fail to Reject RWD Negative 0.209 3.666* 6.742* Reject 10-Month RW Negative 0.228 0.399 0.106 Fail to Reject RWD Negative 0.200 3.685* 6.810* Reject 11-Month RW Negative 0.363 0.117 0.073 Fail to Reject RWD Negative 0.357 3.308* 5.536* Reject 12-Month RW Negative 0.312 -0.371 0.117 Fail to Reject RWD Negative 0.255 2.717* 3.725* Reject 13-Month RW Negative 0.395 -0.730 0.345 Fail to Reject RWD Negative 0.340 2.310* 2.725* Reject 14-Month RW Negative 0.350 -0.801 0.382 Fail to Reject RWD Negative 0.235 2.147* 2.332 Reject 15-Month RW Negative 0.393 -0.821 0.415 Fail to Reject RWD Negative 0.344 2.220* 2.524* Reject 16-Month RW Negative 0.853 -1.260 1.157 Fail to Reject RWD Positive -0.631 1.829* 1.871 Reject 17-Month RW Negative 0.932 -1.087 1.025 Fail to Reject RWD Negative 0.608 1.681* 1.598 Reject 18-Month RW Negative 1.128 -0.965 1.101 Fail to Reject RWD Negative 0.789 1.809* 1.948 Reject 19-Month RW Negative 1.132 -0.650 0.852 Fail to Reject RWD Negative 0.657 1.962* 2.141 Reject 20-Month RW Negative 1.419* -0.023 1.007 Fail to Reject RWD Negative 1.040 2.489* 3.638* Reject

38 RW Negative 1.655* 0.186 1.387 Fail to Reject 21-Month RWD Negative 1.176 2.671* 4.258* Reject 22-Month RW Negative 1.688* -0.224 1.450 Fail to Reject RWD Negative 1.173 2.233* 3.181* Reject 23-Month RW Negative 1.830* -0.783 1.982 Fail to Reject RWD Negative 1.288 1.625* 2.149 Reject RW Negative 1.580* -0.999 1.748 Fail to Reject 24-Month RWD Negative 1.003 1.253 1.289 Fail to Reject † 퐵퐾 퐿푇퐹 퐻표: 푀푆퐸(푒 ) = 푀푆퐸(푒 ), where MSE is the mean square error of benchmark and LTF forecast errors.

One-month-ahead LTF forecasts are significantly more accurate than the RW forecasts.

After that, differences in accuracy are not statistically significant at any other step-length, including those for 12- to 24-months when the RW errors are slightly smaller (Table 6). The LTF forecast errors are significantly smaller than those of the RWD models at all but the longest step- length. Favorable LTF forecasting outcomes for the Paso del Norte bridge may result from adopting a measure of the total cost of crossing that involves both time and physical tolls. A study by Fullerton et al. (2013) uses tolls and other explanatory variables, but omits wait times.

Indeterminate forecasting results in that effort imply that tolls alone may not adequately measure travel costs across this POE.

39 Table 9. Ysleta-Zaragoza northbound small vehicles U-statistics

Step- Observations LTF RW RWD Length 1-Month 48 0.028 0.044 0.070 2-Month 47 0.031 0.044 0.071 3-Month 46 0.029 0.044 0.071 4-Month 45 0.026 0.044 0.071 5-Month 44 0.025 0.045 0.072 6-Month 43 0.026 0.040 0.069 7-Month 42 0.026 0.040 0.066 8-Month 41 0.026 0.033 0.063 9-Month 40 0.025 0.033 0.060 10-Month 39 0.026 0.032 0.059 11-Month 38 0.027 0.032 0.058 12-Month 37 0.027 0.032 0.058 13-Month 36 0.028 0.032 0.058 14-Month 35 0.027 0.032 0.058 15-Month 34 0.029 0.032 0.058 16-Month 33 0.030 0.031 0.057 17-Month 32 0.031 0.032 0.057 18-Month 31 0.031 0.032 0.058 19-Month 30 0.031 0.032 0.059 20-Month 29 0.031 0.033 0.059 21-Month 28 0.030 0.033 0.060 22-Month 27 0.032 0.033 0.061 23-Month 26 0.033 0.034 0.061 24-Month 25 0.034 0.034 0.061

Table 9 reports the U-statistics for the Ysleta-Zaragoza bridge simulations. The LTF U- statistics are smaller than RW U-statistics for all step-lengths other than that for 24-months-ahead.

The LTF and RW U-statistics are lower than those from the RWD model across all 24 step-lengths.

40 Table 10. Ysleta-Zaragoza U-statistic decompositions

LTF RW RWD Step- Bias Var Covar Bias Var Covar Bias Var Covar Length 1-Month 0.001 0.000 0.999 0.009 0.001 0.990 0.000 0.132 0.868 2-Month 0.000 0.001 0.999 0.006 0.000 0.994 0.000 0.132 0.868 3-Month 0.001 0.001 0.999 0.006 0.000 0.994 0.000 0.145 0.855 4-Month 0.011 0.002 0.988 0.004 0.000 0.996 0.000 0.140 0.860 5-Month 0.021 0.000 0.979 0.004 0.000 0.996 0.000 0.147 0.853 6-Month 0.006 0.066 0.928 0.022 0.027 0.951 0.002 0.249 0.749 7-Month 0.001 0.275 0.724 0.022 0.037 0.941 0.000 0.174 0.826 8-Month 0.005 0.261 0.734 0.005 0.000 0.994 0.003 0.125 0.872 9-Month 0.019 0.280 0.701 0.004 0.001 0.995 0.000 0.116 0.884 10-Month 0.025 0.251 0.724 0.014 0.002 0.984 0.001 0.150 0.849 11-Month 0.027 0.322 0.651 0.009 0.000 0.991 0.000 0.111 0.889 12-Month 0.032 0.355 0.612 0.010 0.000 0.990 0.000 0.128 0.872 13-Month 0.025 0.387 0.587 0.016 0.005 0.978 0.000 0.157 0.843 14-Month 0.032 0.415 0.553 0.009 0.000 0.991 0.000 0.129 0.871 15-Month 0.028 0.424 0.549 0.014 0.004 0.982 0.000 0.172 0.828 16-Month 0.033 0.447 0.519 0.006 0.000 0.994 0.000 0.144 0.856 17-Month 0.037 0.399 0.564 0.006 0.000 0.994 0.000 0.153 0.847 18-Month 0.053 0.394 0.553 0.005 0.000 0.995 0.000 0.154 0.846 19-Month 0.066 0.398 0.535 0.008 0.000 0.992 0.000 0.168 0.832 20-Month 0.079 0.397 0.524 0.005 0.000 0.995 0.000 0.158 0.842 21-Month 0.099 0.399 0.502 0.006 0.000 0.994 0.000 0.166 0.834 22-Month 0.102 0.322 0.575 0.009 0.000 0.990 0.000 0.181 0.819 23-Month 0.108 0.330 0.562 0.005 0.000 0.995 0.000 0.168 0.832 24-Month 0.093 0.348 0.559 0.010 0.000 0.989 0.001 0.195 0.804

Second-moment U-statistic decompositions in Table 10 follow the same general patterns as those for the Paso del Norte summary in Table 7. Though forecasts for each of the three models are essentially unbiased, the LTF model struggles to replicate some of the variability for forecasts beyond 6 months in the future. Based on U-statistics alone, the LTF model is more precise than either benchmark, but it may be possible to improve the quality of those forecasts by replicating more of the series variability.

41 Table 11. Ysleta-Zaragoza Ashley et al. (1980) error-differential regression results

Step- Benchmark Error α or α α or α 1 3 2 4 F-statistic Conclusion† Length Forecast Mean t-statistic t-statistic

1-Month RW Negative -0.978 5.040* 13.178* Reject RWD Positive -0.070 8.221* 33.798* Reject 2-Month RW Negative -0.579 2.856* 4.247* Reject RWD Positive 0.054 6.394* 20.444* Reject 3-Month RW Negative -0.297 3.221* 5.230* Reject RWD Negative 0.057 6.868* 23.585* Reject 4-Month RW Negative 0.001 4.287* 9.191* Reject RWD Positive -0.309 8.139* 33.168* Reject 5-Month RW Negative 0.104 4.747* 11.272* Reject RWD Negative 0.416 8.456* 35.839* Reject 6-Month RW Negative -0.510 3.173* 5.165* Reject RWD Negative -0.186 7.528* 28.352* Reject 7-Month RW Negative -0.616 3.081* 4.937* Reject RWD Negative 0.099 6.976* 24.335* Reject 8-Month RW Negative -0.060 1.422* 1.013 Fail to Reject RWD Positive 0.194 6.171* 19.059* Reject 9-Month RW Negative 0.169 1.859* 1.742 Reject RWD Negative 0.417 6.278* 19.797* Reject 10-Month RW Negative 0.041 1.408* 0.992 Reject RWD Negative 0.265 5.967* 17.839* Reject 11-Month RW Negative 0.179 1.184 0.717 Fail to Reject RWD Positive -0.530 5.370* 14.557* Reject 12-Month RW Negative 0.204 1.224 0.770 Fail to Reject RWD Negative 0.493 5.426* 14.840* Reject 13-Month RW Negative 0.044 0.907 0.412 Fail to Reject RWD Negative 0.351 5.066* 12.892* Reject 14-Month RW Negative 0.225 0.916 0.445 Fail to Reject RWD Positive -0.539 4.908* 12.188* Reject 15-Month RW Negative 0.126 0.490 0.128 Fail to Reject RWD Negative 0.385 4.352* 9.545* Reject 16-Month RW Negative 0.348 0.327 0.114 Fail to Reject RWD Positive -0.509 4.074* 8.429* Reject 17-Month RW Negative 0.385 0.192 0.092 Fail to Reject RWD Negative 0.526 3.880* 7.667* Reject 18-Month RW Negative 0.543 0.205 0.169 Fail to Reject RWD Negative 0.655 3.772* 7.327* Reject 19-Month RW Negative 0.552 0.257 0.186 Fail to Reject RWD Negative 0.633 3.838* 7.565* Reject 20-Month RW Negative 0.662 0.441 0.316 Fail to Reject RWD Positive -0.956 3.980* 8.376* Reject

42 RW Negative 0.712 0.704 0.502 Fail to Reject 21-Month RWD Negative 0.806 4.251* 9.359* Reject 22-Month RW Negative 0.655 0.469 0.324 Fail to Reject RWD Negative 0.766 3.965* 8.155* Reject 23-Month RW Negative 0.801 0.320 0.372 Fail to Reject RWD Positive -1.010 3.613* 7.039* Reject RW Negative 0.617 0.180 0.206 Fail to Reject 24-Month RWD Negative 0.678 3.427* 6.102* Reject † 퐵퐾 퐿푇퐹 퐻표: 푀푆퐸(푒 ) = 푀푆퐸(푒 ), where MSE is the mean square error of benchmark and LTF forecast errors.

Results from Ashley, et al. (1980) error-differential regression tests at the Ysleta-Zaragoza

POE are arranged in Table 11. This analysis determines that LTF models are significantly more accurate than RW models at steps 1 through 7 as well as at steps 9 and 10. At longer step-lengths, error-differential regression tests are unable to determine a significant difference between either forecast. Table 10 illustrates this LTF inefficiency through large LTF variance components beyond the 6th step-length. These results contrast with the Paso del Norte findings, where a statistically significant difference exists only at a single step. Error differences between LTF and

RWD models at Paso del Norte and Ysleta-Zaragoza are similar, nonetheless. Against the RWD benchmark, significant differences are present at every step, indicating that the LTF forecasts are significantly more accurate than RWD forecasts. It is difficult to distinguish between the LTF and

RW approaches in terms of accuracy at this bridge. This stands in contrast to the results for the

Ysleta-Zaragoza analysis completed by Fullerton et al. (2013) where the predictive superiority of the LTF model is established at every step.

43 Table 12. Bridge of the Americas northbound small vehicles U-statistics

Step- Observations LTF RW RWD Length 1-Month 48 0.025 0.044 0.077 2-Month 47 0.027 0.044 0.078 3-Month 46 0.028 0.044 0.078 4-Month 45 0.030 0.044 0.077 5-Month 44 0.032 0.044 0.077 6-Month 43 0.033 0.044 0.078 7-Month 42 0.033 0.045 0.079 8-Month 41 0.031 0.045 0.079 9-Month 40 0.030 0.046 0.080 10-Month 39 0.031 0.046 0.081 11-Month 38 0.031 0.047 0.082 12-Month 37 0.030 0.047 0.083 13-Month 36 0.030 0.048 0.084 14-Month 35 0.030 0.048 0.084 15-Month 34 0.031 0.048 0.085 16-Month 33 0.030 0.046 0.083 17-Month 32 0.032 0.047 0.083 18-Month 31 0.035 0.047 0.084 19-Month 30 0.035 0.048 0.084 20-Month 29 0.037 0.049 0.085 21-Month 28 0.038 0.049 0.087 22-Month 27 0.039 0.050 0.088 23-Month 26 0.038 0.051 0.090 24-Month 25 0.038 0.052 0.091

Bridge of the Americas out-of-sample simulation U-statistics are shown in Table 12. The

LTF model yields smaller values than either benchmark at each of the 24 steps. Bridge of the

Americas RW U-statistics are also smaller than those from the RWD model. Among the three bridges for which the personal vehicle traffic flows are analyzed, the LTF forecasts perform best for the Bridge of the Americas in terms of U-statistic comparisons with random walk benchmarks.

44 Table 13. Bridge of the Americas U-statistic decompositions

LTF RW RWD Step- Bias Var Covar Bias Var Covar Bias Var Covar Length 1-Month 0.000 0.066 0.934 0.000 0.000 1.000 0.000 0.207 0.793 2-Month 0.001 0.043 0.956 0.000 0.000 1.000 0.000 0.213 0.787 3-Month 0.000 0.035 0.964 0.001 0.004 0.995 0.000 0.246 0.754 4-Month 0.000 0.029 0.970 0.000 0.000 1.000 0.000 0.221 0.779 5-Month 0.000 0.032 0.968 0.000 0.000 1.000 0.000 0.227 0.773 6-Month 0.002 0.034 0.963 0.000 0.000 1.000 0.000 0.230 0.770 7-Month 0.003 0.034 0.963 0.000 0.000 1.000 0.000 0.236 0.764 8-Month 0.007 0.066 0.927 0.000 0.000 1.000 0.000 0.233 0.767 9-Month 0.005 0.079 0.916 0.000 0.000 1.000 0.000 0.234 0.766 10-Month 0.004 0.124 0.872 0.000 0.000 1.000 0.000 0.237 0.763 11-Month 0.005 0.127 0.867 0.000 0.000 1.000 0.000 0.236 0.764 12-Month 0.006 0.162 0.832 0.000 0.000 1.000 0.000 0.238 0.761 13-Month 0.006 0.154 0.840 0.000 0.000 1.000 0.000 0.238 0.762 14-Month 0.006 0.150 0.844 0.000 0.000 1.000 0.000 0.244 0.756 15-Month 0.015 0.133 0.853 0.001 0.003 0.996 0.001 0.269 0.731 16-Month 0.015 0.147 0.838 0.000 0.000 1.000 0.000 0.245 0.754 17-Month 0.015 0.171 0.814 0.000 0.000 1.000 0.000 0.244 0.756 18-Month 0.014 0.117 0.869 0.001 0.000 0.999 0.000 0.243 0.757 19-Month 0.016 0.116 0.868 0.000 0.000 1.000 0.000 0.247 0.752 20-Month 0.020 0.105 0.874 0.000 0.000 1.000 0.000 0.244 0.756 21-Month 0.016 0.109 0.875 0.001 0.000 0.999 0.000 0.245 0.755 22-Month 0.012 0.138 0.850 0.001 0.000 0.999 0.000 0.242 0.757 23-Month 0.013 0.140 0.847 0.001 0.000 0.999 0.000 0.243 0.757 24-Month 0.019 0.159 0.822 0.000 0.000 1.000 0.001 0.246 0.753

Bias, variance, and covariance second-moment decompositions for Bridge of the Americas

U-statistics are in Table 13. Similar to results from the Paso del Norte and Ysleta-Zaragoza bridges, LTF, RW, and RWD forecasts for the Bridge of the Americas are unbiased, but LTF and

RW forecasts have a smaller variance decomposition than the RWD model. After 9-months-ahead,

LTF forecasts do not capture variability as well as RW forecasts. The disparity in variance components between the two is small, however, with a difference of less than 0.2 at step-lengths larger than 9-months-ahead.

45 Table 14. Bridge of the Americas Ashley et al. (1980) error-differential regression results

Step- Benchmark Error α or α α or α 1 3 2 4 F-statistic Conclusion† Length Forecast Mean t-statistic t-statistic

1-Month RW Negative -0.027 5.128* 13.151* Reject RWD Negative 0.032 10.034* 50.338* Reject 2-Month RW Negative -0.007 3.977* 7.908* Reject RWD Negative 0.011 8.793* 38.662* Reject 3-Month RW Negative -0.139 3.352* 5.628* Reject RWD Negative -0.123 8.149* 33.209* Reject 4-Month RW Negative 0.035 2.742* 3.759* Reject RWD Positive 0.050 7.317* 26.769* Reject 5-Month RW Negative -0.003 2.233* 2.493* Reject RWD Negative -0.046 6.536* 21.362* Reject 6-Month RW Negative 0.199 2.121* 2.269 Reject RWD Negative 0.108 6.341* 20.110* Reject 7-Month RW Negative 0.192 2.165* 2.362 Reject RWD Negative 0.112 6.366* 20.268* Reject 8-Month RW Positive -0.257 2.697* 3.670* Reject RWD Positive -0.260 7.009* 24.599* Reject 9-Month RW Positive -0.191 2.912* 4.258* Reject RWD Negative 0.194 7.231* 26.165* Reject 10-Month RW Negative 0.339 2.850* 4.119* Reject RWD Negative 0.128 7.050* 24.860* Reject 11-Month RW Positive -0.187 2.941* 4.342* Reject RWD Negative 0.226 7.121* 25.379* Reject 12-Month RW Positive -0.221 3.235* 5.256* Reject RWD Negative 0.159 7.468* 27.897* Reject 13-Month RW Positive -0.146 3.291* 5.427* Reject RWD Negative 0.252 7.492* 28.095* Reject 14-Month RW Negative 0.352 3.174* 5.098* Reject RWD Negative 0.081 7.276* 26.476* Reject 15-Month RW Negative 0.328 3.059* 4.734* Reject RWD Negative 0.167 7.221* 26.086* Reject 16-Month RW Positive -0.274 2.846* 4.087* Reject RWD Positive -0.193 6.947* 24.149* Reject 17-Month RW Positive -0.316 2.654* 3.571* Reject RWD Negative 0.268 6.497* 21.140* Reject 18-Month RW Positive -0.237 2.041* 2.110 Reject RWD Positive -0.322 5.697* 16.278* Reject 19-Month RW Positive -0.334 1.909* 1.878 Reject RWD Negative 0.233 5.485* 15.068* Reject 20-Month RW Positive -0.368 1.726* 1.558 Reject RWD Negative 0.388 5.206* 13.629* Reject

46 RW Positive -0.245 1.537* 1.212 Reject 21-Month RWD Positive -0.325 4.946* 12.285* Reject 22-Month RW Positive -0.192 1.510* 1.158 Reject RWD Negative 0.248 4.843* 11.756* Reject 23-Month RW Positive -0.189 1.634* 1.354 Reject RWD Negative 0.263 4.948* 12.276* Reject RW Positive -0.323 1.754* 1.590 Reject 24-Month RWD Negative 0.241 5.078* 12.924* Reject † 퐵퐾 퐿푇퐹 퐻표: 푀푆퐸(푒 ) = 푀푆퐸(푒 ), where MSE is the mean square error of benchmark and LTF forecast errors.

Ashley et al. (1980) error-differential regression results for the Bridge of the Americas are presented in Table 14. Unlike forecast simulation results from the tolled Paso del Norte and

Ysleta-Zaragoza bridges, the analysis of the untolled Bridge of the Americas indicates that LTF forecasts are reliably more accurate than those of the RW and RWD models. Past studies regarding traffic volumes through POEs in El Paso have focused on the effect of tolls and either omit the

Bridge of the Americas as a point of study or do not include an error-differential analysis.

Table 15 consolidates model optimality results from the preceding tables. LTF U-statistics and Ashley et al. (1980) results are contrasted against the two benchmarks at all three POEs.

Figures in the RW column indicate which model performs best when comparing LTF and RW simulation accuracy at each step-length. Figures under the RWD heading reflect comparisons between LTF and RWD models. An octothorpe (#) in the U-statistic columns signals that LTF and benchmark accuracy is equivalent at that step. A hyphen (-) under an AGS heading means that the LTF forecast is not a significant improvement over the benchmark forecast at that step.

47 Table 15. U-statistic and Ashley et al. (1980) accuracy comparisons

Paso del Norte Ysleta-Zaragoza Bridge of the Americas U-statistic AGS U-statistic AGS U-statistic AGS Step RW RWD RW RWD RW RWD RW RWD RW RWD RW RWD 1 LTF LTF LTF LTF LTF LTF LTF LTF LTF LTF LTF LTF 2 LTF LTF - LTF LTF LTF LTF LTF LTF LTF LTF LTF 3 LTF LTF - LTF LTF LTF LTF LTF LTF LTF LTF LTF 4 LTF LTF - LTF LTF LTF LTF LTF LTF LTF LTF LTF 5 LTF LTF - LTF LTF LTF LTF LTF LTF LTF LTF LTF 6 LTF LTF - LTF LTF LTF LTF LTF LTF LTF LTF LTF 7 # LTF - LTF LTF LTF LTF LTF LTF LTF LTF LTF 8 # LTF - LTF LTF LTF - LTF LTF LTF LTF LTF 9 LTF LTF - LTF LTF LTF LTF LTF LTF LTF LTF LTF 10 LTF LTF - LTF LTF LTF LTF LTF LTF LTF LTF LTF 11 LTF LTF - LTF LTF LTF - LTF LTF LTF LTF LTF 12 RW LTF - LTF LTF LTF - LTF LTF LTF LTF LTF 13 RW LTF - LTF LTF LTF - LTF LTF LTF LTF LTF 14 RW LTF - LTF LTF LTF - LTF LTF LTF LTF LTF 15 RW LTF - LTF LTF LTF - LTF LTF LTF LTF LTF 16 RW LTF - LTF LTF LTF - LTF LTF LTF LTF LTF 17 RW LTF - LTF LTF LTF - LTF LTF LTF LTF LTF 18 RW LTF - LTF LTF LTF - LTF LTF LTF LTF LTF 19 RW LTF - LTF LTF LTF - LTF LTF LTF LTF LTF 20 RW LTF - LTF LTF LTF - LTF LTF LTF LTF LTF 21 RW LTF - LTF LTF LTF - LTF LTF LTF LTF LTF 22 RW LTF - LTF LTF LTF - LTF LTF LTF LTF LTF 23 RW LTF - LTF LTF LTF - LTF LTF LTF LTF LTF 24 RW LTF - - # LTF - LTF LTF LTF LTF LTF

At the Paso del Norte bridge, the RW model is highly competitive with the LTF. At that crossing, the LTF approach is a better alternative for most of the shorter forecast horizons.

However, Ashley et al. (1980) results indicate that LTF forecasts are not significantly more accurate than RW forecasts beyond the initial step. At the Ysleta-Zaragoza bridge, U-statistics determine that the LTF methodology is superior to the benchmark models in all but the final step, where the accuracy of the LTF forecast is equal to that of RW forecasts. Ashley et al. (1980) error- 48 differential tests confirm significant differences between LTF and RW forecast accuracy for nine of the first 10 simulations, but significance in error differences cannot be established after that.

For the Bridge of the Americas, U-statistics and Ashley et al. (1980) assessments establish that the

LTF model is a significant improvement over the benchmark models at each step. Additionally,

U-statistic and Ashley et al. (1980) methods determine that LTF equations at each POE yield better accuracy than RWD models in all but one case. The single exception to this pattern occurs for the final step-length for the Paso del Norte bridge. Overall, the information in Table 15 confirms substantial predictive superiority for the LTF equations estimated for these three POEs.

Incorporating border wait times as an independent variable to model northbound traffic volumes across El Paso POEs produces mixed results when compared to the Fullerton et al. study from 2013. For the Paso del Norte analysis, the inclusion of wait times improves forecast accuracy against a RW benchmark as determined by both U-statistic and error-differential assessment. U- statistics results for the Ysleta-Zaragoza bridge are similar in both studies, but error-differential tests suggest that the forecasting performance of the 2013 LTF model is superior. Those outcomes imply that the Ysleta-Zaragoza model may be better-specified without a measure of time cost.

Wait times are also included for the Bridge of the Americas forecast. Though forecasting results overwhelmingly favor the LTF model at this POE, there are no upstream studies by which to assess the effect of wait times on forecast precision.

49 Chapter 6: Conclusion

Border wait times have been shown to impose a negative effect on the economic activity of some border cities (Ghaddar et al., 2004). Long wait times imply increased time costs which, in conjunction with fuel usage and tolls, contribute to the overall burden of crossing. While previous studies have reported an inelastic response to tolls at El Paso POEs, the effect of wait times across these international crossings has not been studied very extensively.

Three LTFs are developed to analyze northbound personal vehicle traffic volumes from

Ciudad Juarez, Mexico to El Paso, Texas. Each equation applies to one of the bridges included in the study: Paso del Norte, Ysleta-Zaragoza, or Bridge of the Americas. Sample data include real tolls, wait times at the same bridge, wait times at the other bridges, business cycle fluctuations in

Ciudad Juarez and El Paso, the real peso-to-dollar exchange rate, and the Mexican industrial production index.

Overall, toll hikes negatively influence traffic at the Paso del Norte and Ysleta-Zaragoza

POEs, but increase volumes across the Bridge of the Americas. That occurs because the latter structure is untolled and serves as a substitute route for the tolled bridges. Increased wait times at any of the three bridges are linked to decreased traffic volumes through those POEs, while simultaneously increasing traffic flows at the other two crossings. An increase in either IMMEX employment in Ciudad Juarez, or non-agricultural employment in El Paso, produces a positive effect on crossings at all three bridges. Increases in the real exchange rate have negative effects on bridge traffic flows that may result from decreased spending power for Mexican consumers following a peso depreciation. At each POE, personal vehicle traffic volumes escalate in response

50 to gains in industrial production in Mexico, as well. Wait time toll-equivalent estimates at the

Paso del Norte and Ysleta-Zaragoza crossings are estimated at around $0.04 per minute, or $2.40 per hour of wait time. At the Bridge of the Americas, the toll-equivalent estimate is $0.03, or

$1.80 per hour of wait time. The 2018 nominal hourly time costs are $2.50 and $1.88, respectively.

With long wait times, the cost of crossing at a specific POE can become prohibitive. Using easily accessible real-time sources to monitor wait times throughout the El Paso area, border crossers can make educated choices between structures to reduce the cost burden. Distances between bridges in Ciudad Juarez, traffic congestion, and extended drives from the alternative crossing to the El Paso destination contribute to transactional costs which also influence commuter

POE selection. If an alternate route is selected, the total cost of using the substitute POE will generally be lower than the crossing cost at the preferred selection.

The fidelity of the estimation results is examined by comparing the accuracy of out-of- sample LTF forecasts against RW and RWD benchmark forecasts. In general, U-statistics and error-differential regression statistics indicate that LTF equations achieve superior predictive accuracy relative to either benchmark, but caution should be exercised in interpreting these results.

The inclusion of wait times as a regressor for the Paso del Norte LTF model appears to improve northbound personal vehicle crossing forecast accuracy. The same improvement is not apparent for the Ysleta-Zaragoza bridge. The effect of wait times on forecast precision for the Bridge of the Americas is indeterminate because LTF forecasts have not previously been assembled for this structure.

51 Consumers respond to price and income differentials across international boundaries.

Because these differentials are often large between the United States and Mexico, it would be helpful to examine the cost and effect of border wait times at POEs between countries with more similar economic profiles, such as those in western Europe. Replication of this study for pedestrian traffic flows across the bridges used in this study may help shed light on substitution effects between driving and walking. The methodology can also be applied to any other region where infrastructure and/or administrative bottlenecks create wait times that potentially affect commerce.

52 References

Ashley, R., Granger, C. W. J., & Schmalensee, R. (1980). Advertising and aggregate consumption: an analysis of causality. Econometrica, 48(5): 1149-1167. Baruca, A., & Zolfagharian, M. (2013). Cross-border shopping: Mexican shoppers in the US and American shoppers in Mexico. International Journal of Consumer Studies, 37(4): 360- 366. BRMP. (2017). Southbound International Bridge Tolls. Unpublished report. El Paso, TX: University of Texas at El Paso Border Region Modeling Project. BTS. (2016). Border Crossing/Entry Data. Washington, DC: U.S. Bureau of Transportation Statistics. CAPUFE. (2017) Tarifas. Cuernavaca, Morelos: MX. Caminos y Puentes Federales de Ingresos y Servicios Conexos. CBP. (2013). CBP Partners with Private Sector on Port of Entry Service Reimbursement. Washington, DC: U.S. Customs and Border Protection. CBP. (2017). Border Wait Times. Washington, DC: U.S. Customs and Border Protection. Coronado, R. A., & Phillips, K. R. (2007). Exported retail sales along the Texas-Mexico border. Journal of Borderlands Studies, 22(1): 19-38. Del Castillo-Vera, G. (2009). Tiempos de espera en los cruces fronterizos del norte de México: Una barrera no arancelaria. Comercio Exterior, 59(7): 551-557. De Leon, M., Fullerton Jr, T. M., & Kelley, B. W. (2009) Tolls, exchange rates, and borderplex international bridge traffic. International Journal of Transport Economics, 36(2): 223- 259. De Vany, A. (1974). The revealed value of time in air travel. Review of Economics and Statistics, 56(1): 77-82. Earnhart, D. (2004). Time is money: improved valuation of time and transportation costs. Environmental and Resource Economics, 29(2): 159-190. ERB. (2008). Economic Impact Study: Otay Mesa East Port of Entry. La Jolla, CA: Export Access. Fullerton Jr, T. M. (2000). Currency movements and international border crossings. International Journal of Public Administration, 23(5-8): 1113-1123. Fullerton, T. M., & Kelley, B. (2008). El Paso housing sector econometric forecast accuracy. Journal of Agricultural and Applied Economics, 40(1): 385-402. Fullerton, T. M., Molina, A. L., & Walke, A. G. (2013). Tolls, exchange rates and northbound international bridge traffic from Mexico. Regional Science Policy & Practice, 5(3): 305- 321. FRED. (2017). Economic Data. St. Louis, MO: Federal Reserve Bank of St. Louis.

53 GAO. (2013). CBP Action Needed to Improve Wait Time Data and Measure Outcomes of Trade Facilitation Efforts. Washington, DC: Government Accountability Office. Ghaddar, S., Richardson, C., & Brown, C. J. (2004). The Economic Impact of Mexican Visitors to the Lower Rio Grande Valley 2003. Edinburg, TX: University of Texas-Pan American Center for Border Economic Studies. Hensher, D. A., Chinh, H.Q., & Wen, L. (2016). How much is too much for tolled road users: toll saturation and the implications for car commuting value of travel time savings? Transportation Research Part A: Policy and Practice, 94: 604-621. Hirschman, I., Mcknight, C., Pucher, J., Paaswell, R. E., & Berechman, J. (1995). Bridge and tunnel toll elasticities in New York: Some recent evidence. Transportation, 22(2): 97- 113. INEGI. (2017). Sistema Estatal y Municipal de Bases de Datos. Aguascalientes, Aguascalientes: MX. Instituto Nacional de Estadística y Geografía. Matas, A., & Raymond, J. L. (2003). Demand elasticity on tolled motorways. Journal of Transportation and Statistics, 6(2/3): 91-108. McConnell, K. E., & Strand, I. (1981). Measuring the cost of time in recreation demand analysis: An application to sportfishing. American Journal of Agricultural Economics, 63(1): 153- 156. Olmedo, C. & Soden, D.L. (2005). Terrorism's role in re-shaping border crossings: 11 September and the US borders. Geopolitics, 10(4): 741-766 Pindyck, R. S., & Rubinfeld, D. L. (1998). Econometric Models and Econometric Forecasts (4th ed.). Boston, MA: Irwin McGraw-Hill. Roberts, B., Rose, A., Heatwole, N., Wei, D., Avetisyan, M., Chan, O., & Maya, I. (2014). The impact on the US economy of changes in wait times at ports of entry. Transport Policy, 35: 162-175. Small, K, A. (2013). Urban transportation economics. In R. J. Arnott (1st ed.), Regional and Urban Economics: Part 1 (pp. 270-459). New York, NY: Routledge. TTI. (2017). How Long is too Long to Cross the Border? College Station, TX: Texas A&M Transportation Institute.

54 Data Appendix

Table 16. USCPI, NEXR, and Northbound Border Crossings Data

USCPI BCPN BCYN BCAN NEXR 1982-1984=100 (millions) (millions) (millions) Aug-10 217.923 12.766 0.300 0.197 0.187 Sep-10 218.275 12.798 0.285 0.193 0.169 Oct-10 219.035 12.439 0.301 0.201 0.176 Nov-10 219.590 12.338 0.287 0.193 0.163 Dec-10 220.472 12.390 0.296 0.205 0.165 Jan-11 221.187 12.128 0.289 0.198 0.150 Feb-11 221.898 12.065 0.248 0.165 0.136 Mar-11 223.046 11.996 0.278 0.187 0.157 Apr-11 224.093 11.706 0.270 0.176 0.150 May-11 224.806 11.654 0.267 0.182 0.149 Jun-11 224.806 11.806 0.259 0.180 0.151 Jul-11 225.395 11.674 0.277 0.190 0.160 Aug-11 226.106 12.237 0.293 0.196 0.168 Sep-11 226.597 13.064 0.279 0.186 0.159 Oct-11 226.750 13.438 0.278 0.187 0.163 Nov-11 227.169 13.696 0.252 0.162 0.155 Dec-11 227.223 13.775 0.278 0.164 0.160 Jan-12 227.842 13.383 0.285 0.172 0.170 Feb-12 228.329 12.783 0.264 0.157 0.157 Mar-12 228.807 12.752 0.283 0.168 0.178 Apr-12 229.187 13.056 0.267 0.157 0.173 May-12 228.713 13.620 0.266 0.163 0.173 Jun-12 228.524 13.919 0.260 0.164 0.177 Jul-12 228.590 13.364 0.272 0.173 0.181 Aug-12 229.918 13.179 0.278 0.186 0.184 Sep-12 231.015 12.924 0.267 0.173 0.186 Oct-12 231.638 12.898 0.280 0.177 0.194 Nov-12 231.249 13.064 0.275 0.176 0.191 Dec-12 231.221 12.865 0.284 0.199 0.207 Jan-13 231.612 12.696 0.285 0.172 0.233 Feb-13 232.985 12.725 0.256 0.163 0.230 Mar-13 232.299 12.500 0.299 0.191 0.251 Apr-13 231.795 12.206 0.297 0.177 0.248 May-13 231.916 12.299 0.296 0.194 0.173 Jun-13 232.374 12.964 0.290 0.196 0.177 Jul-13 232.889 12.762 0.309 0.208 0.263 Aug-13 233.323 12.912 0.313 0.209 0.267 Sep-13 233.632 13.055 0.305 0.197 0.232

55 Oct-13 233.718 12.992 0.315 0.207 0.252 Nov-13 234.121 13.060 0.308 0.200 0.249 Dec-13 234.723 13.010 0.323 0.217 0.232 Jan-14 235.385 13.222 0.302 0.200 0.255 Feb-14 235.672 13.293 0.270 0.176 0.242 Mar-14 235.978 13.193 0.322 0.224 0.271 Apr-14 236.471 13.067 0.311 0.215 0.273 May-14 236.832 12.933 0.330 0.242 0.278 Jun-14 237.029 12.993 0.310 0.218 0.267 Jul-14 237.424 12.991 0.312 0.225 0.278 Aug-14 237.256 13.144 0.331 0.223 0.276 Sep-14 237.486 13.237 0.330 0.210 0.266 Oct-14 237.506 13.480 0.331 0.234 0.282 Nov-14 237.118 13.615 0.313 0.229 0.267 Dec-14 236.290 14.521 0.352 0.224 0.285 Jan-15 234.913 14.697 0.336 0.194 0.270 Feb-15 235.489 14.917 0.302 0.224 0.257 Mar-15 235.989 15.238 0.320 0.261 0.287 Apr-15 236.201 15.194 0.313 0.253 0.273 May-15 236.891 15.280 0.325 0.246 0.294 Jun-15 237.419 15.479 0.315 0.253 0.277 Jul-15 237.876 15.952 0.321 0.247 0.289 Aug-15 237.811 16.534 0.317 0.257 0.297 Sep-15 237.467 16.839 0.313 0.235 0.290 Oct-15 237.792 16.570 0.327 0.245 0.301 Nov-15 238.153 16.631 0.334 0.227 0.298 Dec-15 237.846 17.070 0.338 0.229 0.304 Jan-16 238.106 18.065 0.340 0.222 0.301 Feb-16 237.808 18.433 0.304 0.216 0.283 Mar-16 238.078 17.630 0.348 0.244 0.318 Apr-16 238.908 17.480 0.332 0.234 0.315 May-16 239.362 18.136 0.426 0.230 0.323 Jun-16 239.842 18.654 0.321 0.215 0.277 Jul-16 239.898 18.616 0.324 0.236 0.318 Aug-16 240.389 18.474 0.313 0.248 0.297 Sep-16 241.006 19.244 0.305 0.225 0.316 Oct-16 241.694 18.891 0.316 0.238 0.331 Nov-16 242.199 20.009 0.288 0.214 0.310 Dec-16 242.821 20.499 0.305 0.249 0.317

56 Table 17. Northbound Personal Vehicle Toll data

Paso Del Norte Ysleta-Zaragoza

(Pesos) (Pesos) Aug-10 $23.00 $23.00 Sep-10 $23.00 $23.00 Oct-10 $23.00 $23.00 Nov-10 $23.00 $23.00 Dec-10 $23.00 $23.00 Jan-11 $23.00 $24.00 Feb-11 $23.00 $24.00 Mar-11 $23.00 $24.00 Apr-11 $23.00 $24.00 May-11 $23.00 $24.00 Jun-11 $23.00 $24.00 Jul-11 $23.00 $24.00 Aug-11 $23.00 $24.00 Sep-11 $23.00 $24.00 Oct-11 $23.00 $24.00 Nov-11 $23.00 $24.00 Dec-11 $24.00 $24.00 Jan-12 $24.00 $25.00 Feb-12 $24.00 $25.00 Mar-12 $24.00 $25.00 Apr-12 $24.00 $25.00 May-12 $24.00 $25.00 Jun-12 $24.00 $25.00 Jul-12 $24.00 $25.00 Aug-12 $24.00 $25.00 Sep-12 $24.00 $25.00 Oct-12 $24.00 $25.00 Nov-12 $24.00 $25.00 Dec-12 $24.00 $25.00 Jan-13 $24.00 $25.00 Feb-13 $24.00 $25.00 Mar-13 $24.00 $25.00 Apr-13 $24.00 $25.00 May-13 $24.00 $25.00 Jun-13 $24.00 $25.00 Jul-13 $24.00 $25.00 Aug-13 $24.00 $25.00 Sep-13 $24.00 $25.00 Oct-13 $24.00 $25.00 Nov-13 $24.00 $25.00 Dec-13 $24.00 $25.00

57 Jan-14 $26.00 $26.00 Feb-14 $26.00 $26.00 Mar-14 $26.00 $26.00 Apr-14 $26.00 $26.00 May-14 $26.00 $26.00 Jun-14 $26.00 $26.00 Jul-14 $26.00 $26.00 Aug-14 $26.00 $26.00 Sep-14 $26.00 $26.00 Oct-14 $26.00 $26.00 Nov-14 $26.00 $26.00 Dec-14 $26.00 $26.00 Jan-15 $26.00 $26.00 Feb-15 $26.00 $26.00 Mar-15 $26.00 $26.00 Apr-15 $26.00 $26.00 May-15 $26.00 $26.00 Jun-15 $26.00 $26.00 Jul-15 $26.00 $26.00 Aug-15 $26.00 $26.00 Sep-15 $26.00 $26.00 Oct-15 $26.00 $26.00 Nov-15 $26.00 $26.00 Dec-15 $26.00 $26.00 Jan-16 $26.00 $26.00 Feb-16 $26.00 $26.00 Mar-16 $26.00 $26.00 Apr-16 $26.00 $26.00 May-16 $26.00 $26.00 Jun-16 $26.00 $26.00 Jul-16 $26.00 $26.00 Aug-16 $26.00 $26.00 Sep-16 $26.00 $26.00 Oct-16 $26.00 $26.00 Nov-16 $26.00 $26.00 Dec-16 $26.00 $26.00

58 Table 18. WTPN, WTYN, and WTAN Data

WTPN WTYN WTAN Aug-10 34.348 35.067 37.524 Sep-10 32.542 35.450 39.730 Oct-10 32.231 33.149 37.447 Nov-10 38.737 34.976 39.589 Dec-10 49.245 39.091 46.051 Jan-11 38.301 30.480 40.347 Feb-11 44.293 35.129 45.786 Mar-11 46.503 43.450 47.704 Apr-11 49.116 49.129 49.087 May-11 41.098 48.029 53.506 Jun-11 45.566 41.868 49.275 Jul-11 36.000 40.589 43.863 Aug-11 32.078 28.477 39.484 Sep-11 39.514 34.284 40.667 Oct-11 40.392 36.409 44.194 Nov-11 45.943 34.828 51.125 Dec-11 35.079 31.530 41.200 Jan-12 30.107 29.962 38.459 Feb-12 46.199 39.703 50.168 Mar-12 40.368 33.132 43.079 Apr-12 48.944 37.663 50.389 May-12 36.601 32.494 41.288 Jun-12 31.315 25.631 37.528 Jul-12 35.271 31.010 37.975 Aug-12 23.991 22.659 32.543 Sep-12 21.121 15.451 27.602 Oct-12 21.690 18.611 27.952 Nov-12 27.319 22.000 34.721 Dec-12 35.314 30.878 38.786 Jan-13 24.411 21.250 29.977 Feb-13 26.113 23.729 35.725 Mar-13 31.743 29.274 35.932 Apr-13 28.877 27.948 33.214 May-13 24.879 27.578 34.249 Jun-13 20.000 25.616 28.297 Jul-13 19.910 23.580 30.137 Aug-13 20.385 19.420 29.817 Sep-13 14.132 14.238 27.344 Oct-13 14.380 13.611 28.435 Nov-13 18.703 14.116 27.413 Dec-13 27.143 22.179 33.843

59 Jan-14 12.220 11.424 25.208 Feb-14 15.843 14.708 27.897 Mar-14 17.685 18.827 29.162 Apr-14 23.005 22.419 32.949 May-14 19.208 18.818 29.255 Jun-14 20.067 21.333 29.580 Jul-14 19.739 23.894 29.252 Aug-14 20.021 25.545 27.714 Sep-14 11.831 16.559 23.974 Oct-14 15.200 15.670 26.167 Nov-14 22.646 26.950 31.242 Dec-14 26.043 29.806 32.138 Jan-15 15.702 21.330 24.736 Feb-15 20.866 28.989 31.683 Mar-15 25.281 33.153 36.000 Apr-15 22.928 31.105 34.086 May-15 27.797 31.212 33.437 Jun-15 23.873 33.039 32.819 Jul-15 28.591 28.227 33.976 Aug-15 25.376 27.525 33.382 Sep-15 21.487 24.847 31.857 Oct-15 21.718 24.909 32.335 Nov-15 28.294 23.773 37.356 Dec-15 35.181 30.931 41.667 Jan-16 23.179 18.195 31.895 Feb-16 21.782 23.035 33.676 Mar-16 27.031 23.139 37.526 Apr-16 27.618 23.747 39.316 May-16 37.027 29.773 46.464 Jun-16 34.749 30.703 40.944 Jul-16 30.318 29.843 42.765 Aug-16 27.706 24.885 39.341 Sep-16 30.422 27.281 39.656 Oct-16 30.902 26.187 42.062 Nov-16 33.277 33.620 36.723 Dec-16 30.203 29.966 40.356

60 Table 19. CJIMX, ELPM, REXR, and MXIP Data

CJIMX ELPM REXR MXIP

(thousands) (thousands) 1997=100 2008=100 Aug-10 180.171 277.400 98.440 100.551 Sep-10 179.698 281.700 98.215 99.857 Oct-10 179.327 282.300 94.968 102.858 Nov-10 176.351 284.200 93.508 100.362 Dec-10 178.282 285.400 93.580 99.856 Jan-11 176.824 281.200 91.584 97.106 Feb-11 178.382 281.600 91.273 94.207 Mar-11 179.420 283.100 91.445 102.180 Apr-11 179.126 284.200 89.888 97.703 May-11 181.908 284.400 90.475 101.502 Jun-11 180.401 282.700 91.567 101.523 Jul-11 178.535 280.600 90.180 101.674 Aug-11 177.150 281.900 94.612 103.675 Sep-11 178.454 285.700 100.803 102.620 Oct-11 178.289 283.000 102.912 106.868 Nov-11 177.606 285.400 103.727 104.952 Dec-11 178.185 286.800 103.149 103.140 Jan-12 179.598 282.800 100.251 100.708 Feb-12 182.594 284.900 95.735 99.904 Mar-12 183.568 286.600 96.208 104.985 Apr-12 186.888 287.600 98.976 100.652 May-12 188.730 288.900 103.678 105.882 Jun-12 188.650 285.800 104.980 105.085 Jul-12 186.650 284.100 100.082 105.343 Aug-12 188.028 286.200 98.975 106.406 Sep-12 190.031 288.600 97.140 104.416 Oct-12 192.339 290.100 96.252 108.837 Nov-12 195.271 292.400 96.505 107.284 Dec-12 194.657 292.800 94.525 102.457 Jan-13 196.017 287.000 93.058 101.641 Feb-13 199.320 289.300 93.536 98.320 Mar-13 202.315 290.500 91.647 101.080 Apr-13 202.947 291.100 89.156 103.386 May-13 202.968 292.300 90.395 104.558 Jun-13 199.378 289.100 95.440 102.643 Jul-13 202.388 287.500 94.081 105.480 Aug-13 201.608 290.500 95.051 106.171 Sep-13 201.616 293.300 95.959 102.974 Oct-13 206.301 294.600 94.700 109.447 Nov-13 206.154 297.500 94.213 106.598 Dec-13 205.355 297.600 93.152 103.347

61 Jan-14 208.539 292.900 94.205 103.562 Feb-14 204.340 294.600 94.726 99.974 Mar-14 207.531 295.700 94.462 105.991 Apr-14 209.940 296.300 94.056 103.247 May-14 212.739 297.400 93.629 107.629 Jun-14 214.956 294.700 94.156 105.854 Jul-14 215.202 290.900 93.822 108.801 Aug-14 215.407 293.000 94.409 108.312 Sep-14 217.871 294.300 94.742 107.092 Oct-14 221.369 296.100 95.696 112.749 Nov-14 223.231 298.500 95.436 108.916 Dec-14 224.294 299.700 100.611 107.051 Jan-15 225.228 296.100 101.469 105.102 Feb-15 226.654 297.900 103.300 102.226 Mar-15 230.908 298.800 105.623 108.239 Apr-15 235.631 300.200 106.098 105.031 May-15 236.911 302.000 105.619 106.518 Jun-15 241.867 301.000 107.317 107.246 Jul-15 244.578 298.200 110.327 109.761 Aug-15 247.730 300.300 114.058 109.560 Sep-15 250.754 302.600 115.658 108.890 Oct-15 254.941 305.600 113.009 112.935 Nov-15 253.226 307.700 112.644 108.839 Dec-15 247.675 307.800 114.559 107.249 Jan-16 247.754 304.200 121.200 105.121 Feb-16 250.577 305.700 123.446 104.555 Mar-16 251.778 306.300 118.273 105.879 Apr-16 253.328 309.200 118.122 106.686 May-16 256.705 310.100 123.675 106.928 Jun-16 254.873 307.300 127.357 107.839 Jul-16 256.069 306.800 126.469 108.181 Aug-16 259.037 308.800 125.361 109.905 Sep-16 264.595 311.300 129.750 107.318 Oct-16 263.463 313.600 127.110 111.703 Nov-16 265.006 316.000 134.100 110.496 Dec-16 266.251 316.300 136.198 106.633

62 Vita

Omar Solis was born in El Paso, Texas. He is an alumnus of the University of Texas at El

Paso. In 2007, Omar earned a Bachelor of Science degree in Biology as a Presidential scholar and completed a Bachelor of Science in Psychology in 2011 with cum laude honors. Omar began work toward a Master of Science in Economics in the summer of 2015. In 2016, he was named a James

Foundation Scholar by the Department of Economics & Finance.

Omar has been employed by the university throughout his graduate studies. Initially hired as a Teaching Assistant for the Department of Economics & Finance, he was later re-assigned as a Research Assistant for the Border Region Modeling Project where he has contributed to multiple reports related to the El Paso Borderplex area. During the Summer of 2018, he worked as a full- time Economic Analyst for the Border Region Modeling Project. Omar has been involved with many community outreach organizations, including the University Lions, the American Red Cross, and the El Paso International Music Foundation.

Contact information: [email protected]

This thesis was typed by Omar Solis

63